{ "features_resolved": { "snapshot_any": true, "snapshot_mem": true, "fork": true, "pause_resume": false, "persistent_disk": true, "whole_disk": null, "persist_scope": null, "volumes": true, "auto_stop_idle": false, "long_sessions": true, "custom_image": true, "image_snapshot": true, "image_oci": true, "vm_isolation": false, "docker_inside": true, "nested_virt": false, "root": true, "systemd": false, "browser": false, "desktop": false, "computer_use": false, "code_interpreter": false, "browser_control": null, "desktop_control": null, "browser_and_desktop": false, "anti_bot": false, "captcha_solving": false, "residential_ip": false, "agent_harnesses": false, "ssh": true, "public_ipv4": false, "inbound_https": true, "custom_domain": false, "raw_tcp_inbound": false, "egress_allowlist": true, "open_internet": true, "static_egress_ip": true, "private_network": null, "self_host": true, "byoc": true, "open_source": true, "soc2": true, "hipaa": true, "sso": false, "gpu": true, "arm64": false, "windows": false, "macos": false, "lang_python": true, "lang_node": true, "lang_go": null, "lang_rust": null, "lang_java": null, "gpu_desktop": false, "credential_injection": null, "egress_http_rules": false, "wake_on_request": false, "live_resize": false, "fork_running": true, "memory_snapshot_fork": true, "webhooks": true, "scheduled_wakeups": true, "audit_logs": null, "eu_data_residency": null, "zero_data_retention": null, "scoped_api_keys": null, "spend_limits": null, "mcp_server": true, "usage_api": false, "snapshot_auto": false, "snapshot_on_demand": true, "harness_api": false, "own_agent_api": false, "fast_boot": false, "strong_isolation": true, "ingress_rules": false, "guest_firewall": false, "secret_proxy": false, "secret_proxy_any": false, "volume_attach": true, "volume_shared": true, "sidecar": null, "fuse": false, "kernel_features": false, "burst": null }, "company": { "id": "beam", "checked": "2026-09-30", "company": "Beam (Smartshare, Inc.); formerly Slai", "ownership": { "value": "independent", "parent": null, "source": "https://www.beam.cloud/about", "quote": "© 2026 Smartshare, Inc. All rights reserved." }, "stage": { "value": "seed", "source": "https://techcrunch.com/2022/05/10/slai-seed-round/", "quote": "The company today announced it raised a $3.5 million seed round led by Tiger Global, with additional investment from Y Combinator, Charge Ventures, Uncorrelated Ventures, Twenty Two Ventures and Soma Capital, along with angels including Guy Podjarny and Jason Warner.", "confidence": "high", "note": "Round was raised in 2022 under the company's original name/product, Slai (same founders Eli Mernit and Luke Lombardi, same investor set later listed on beam.cloud/about: Tiger Global, Charge Ventures, Soma Capital, Y Combinator). Slai was rebranded to Beam (beam.cloud) as an open-source serverless/sandbox GPU cloud. No priced round after the 2022 seed was found in ~15 min of research (no TechCrunch/press hits, no blog announcement); YC's own profile still lists the company as a 5-person team, consistent with no larger follow-on round. Beam's own about page additionally shows Alumni Ventures and Hustle Fund logos (and NVIDIA Inception Program membership, which is a partner program, not an investor), which could indicate a later seed extension/additional seed investors not covered by a dedicated press article; treated as part of the same seed stage absent a differently-dated/priced round." }, "accelerators": [ { "name": "Y Combinator", "batch": "Winter 2022", "source": "https://www.ycombinator.com/companies/beam", "quote": "Beam AI-Native Cloud Platform Y Combinator Logo Winter 2022 Active ... Beam Founded: 2021 Batch: Winter 2022 Team Size: 5 Status: Active Location: New York City, NY", "confidence": "high", "via": "research" } ], "total_funding_usd": { "value": 3500000, "status": "confirmed", "basis": "TechCrunch-reported $3.5M seed round (as Slai, 2022); no additional priced rounds found", "source": "https://techcrunch.com/2022/05/10/slai-seed-round/", "quote": "The company today announced it raised a $3.5 million seed round led by Tiger Global" }, "last_round": { "type": "seed", "amount_usd": 3500000, "date": "2022-05", "leads": [ "Tiger Global" ], "status": "confirmed", "source": "https://techcrunch.com/2022/05/10/slai-seed-round/", "quote": "The company today announced it raised a $3.5 million seed round led by Tiger Global, with additional investment from Y Combinator, Charge Ventures, Uncorrelated Ventures, Twenty Two Ventures and Soma Capital, along with angels including Guy Podjarny and Jason Warner." }, "investors": [ "Tiger Global", "Y Combinator", "Charge Ventures", "Soma Capital", "Uncorrelated Ventures", "Twenty Two Ventures", "Alumni Ventures", "Hustle Fund", "Guy Podjarny", "Jason Warner" ], "portfolio_listings": [], "revenue": { "value": null, "status": "unknown" }, "employees": { "value": 5, "source": "https://www.ycombinator.com/companies/beam (\"Founded in 2021 by Luke Lombardi and Eli Mernit, Beam has 5 employees based in New York City.\"; page also shows \"Team Size: 5\")" }, "note": "Beam (beam.cloud) is the rebrand of Slai, an ML-deployment platform founded in 2021 by Eli Mernit and Luke Lombardi, YC W22 (Active per YC directory), same domain/founders confirmed against the YC hint. The only confirmed funding event found is Slai's May-2022 $3.5M seed led by Tiger Global with YC, Charge Ventures, Uncorrelated Ventures, Twenty Two Ventures, Soma Capital and angels Guy Podjarny and Jason Warner (TechCrunch). Beam's own about page lists Tiger Global, Y Combinator, Charge Ventures, Soma Capital, Alumni Ventures and Hustle Fund as backers, and separately touts NVIDIA Inception Program membership (a technology partner program, not a funding round). No later priced round (seed extension, Series A, etc.) or revenue/ARR figure was found in the ~15-minute research window; buyers should treat total funding and stage as likely stale given the company has clearly grown its product surface (GPU cloud, sandboxes, inference) well beyond the original 2022 ML-deployment tool." }, "id": "beam", "name": "Beam", "url": "https://www.beam.cloud/pricing", "category": "agent-sandbox", "isolation": "gvisor", "modes": [ { "key": "on-demand", "label": "Sandbox (CPU), per-ms on allocated cores + RAM", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": false, "features": { "auto_stop_idle": false, "wake_on_request": false, "isolation": "gvisor", "raw_tcp_inbound": false }, "pricing": "resource", "vcpu_h": 0.0675, "ram_gib_h": 0.02304, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": 0, "granularity_s": 0.001, "start_fee": 0, "boot_overhead_s": 0, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": "$0.0000375/s per physical core (= 2 vCPU) -> $0.135/core-h = $0.0675/vCPU-h; RAM $0.0000064/GiB-s = $0.02304/GiB-h. Page example 1 core/8 GiB = $0.319/h. Machine start and image pull not billed. Billed until terminate() or keep_warm_seconds TTL expiry (idle tail billed; -1 = never). June-2026 rate was $0.19/core-h + $0.0202/GB-h. US by default; no other region is listed (security page 2026-10-14). Basis verified 2026-09-28: \"You are only charged when your containers are running.\" (https://docs.beam.cloud/v2/resources/pricing-and-billing). Sandbox keep_warm_seconds is a TTL (sb.update_ttl(300); -1 = \"Manual termination only\") (https://docs.beam.cloud/v2/sandbox/configuration), not an idle detector: auto_stop_idle = false for sandbox modes; serverless containers still scale down after keep_warm idle. CPU/RAM kept alloc.", "perf": { "cpu_runs_s": null, "cold_start_ms": 105.06, "burst_tti_ms": 1323.74, "dax_s": 64.89 }, "perf_source": "ComputeSDK (github.com/computesdk/benchmarks results/): cold 105.06 ms (seq TTI 2026-08-07, runtime \"node\", SDK default size), burst 1,323.74 ms (2026-09-25), dax 64.89 s (2026-09-25, cpu:8 / 16 GiB; guest saw 96 host CPUs). No HPC CPU result." }, { "key": "sandbox-gpu", "label": "Sandbox with GPU (rate not published)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "features": { "auto_stop_idle": false, "wake_on_request": false, "isolation": "gvisor" }, "pricing": "resource", "vcpu_h": 0.189, "ram_gib_h": 0.0198, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": 0, "granularity_s": 0.001, "start_fee": 0, "boot_overhead_s": 0, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "RTX-4090": null, "RTX-5090": null, "H100": null, "A10": null }, "note": "GPU rates NOT published for sandboxes (pricing page 2026-09-28 lists serverless GPU and GPU machines only), so gpu prices are null and this regime is unpriced. Previously assumed = serverless GPU per-second rates (4090 $0.000192/s, 5090 $0.000303/s, H100 $0.000972/s; A10G $0.000292/s from June-2026 blog) plus GPU-attached CPU $0.000105/core-s and RAM $0.0000055/GiB-s. GPU concurrency 5 (Developer) / 50 (Team)." }, { "key": "vm", "label": "Virtual machine (CPU microVM, persistent root, per-second while running)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": false, "features": { "isolation": "microvm", "auto_stop_idle": true, "wake_on_request": true, "pause_resume": true, "persistent_disk": true, "systemd": true, "root": true, "ssh": true, "desktop": true, "docker_inside": true, "gpu": false, "ingress_rules": true, "raw_tcp_inbound": false }, "pricing": "resource", "vcpu_h": 0.004275, "ram_gib_h": 0.0021375, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": 0, "granularity_s": 1, "start_fee": 0, "boot_overhead_s": 0, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": "Pricing page 2026-10-14 (Virtual Machines): \"Persistent Linux VMs for agents. Billed per second while running, with no compute charges while stopped.\" Default 4 vCPU = $0.0171/h, 8 GiB RAM = $0.0171/h, \"50 GiB root disk Included\", rate $0.0342/h ($0.0000095/s); page constants $0.004275/vCPU-h and $0.0021375/GiB-h. Priced per vCPU (not per physical core like sandboxes), although the CLI calls the flag `--cpu CORES` (default 4). Root disk default 50 GiB, max 100Gi (https://docs.beam.cloud/v2/vm/cli). Idle TTL (`--ttl`, idle-action stop or pause) and auto-resume on access are opt-in; unset TTL keeps running and billing. Docs: \"Each VM runs in a CPU microVM with its own kernel, systemd, a durable root filesystem, and stable service URLs.\" (https://docs.beam.cloud/v2/vm/overview). CPU only (\"GPUs | Unsupported\").", "perf": { "cpu_runs_s": null, "cold_start_ms": null, "burst_tti_ms": null, "dax_s": null }, "perf_source": "No benchmark of Beam VMs yet; ComputeSDK numbers are for the container sandbox." }, { "key": "serverless-container", "label": "Serverless function / pod (not a sandbox), 1/3 of sandbox CPU rate", "flags": [ "alt" ], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": false, "features": { "wake_on_request": true, "isolation": "container" }, "pricing": "resource", "vcpu_h": 0.0225, "ram_gib_h": 0.00756, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": 0, "granularity_s": 0.001, "start_fee": 0, "boot_overhead_s": 0, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": "$0.0000125/core-s ($0.045/core-h) + $0.0000021/GiB-s. Billed incl. on_start and keep_warm_seconds (defaults: endpoints 180 s, task queues 10 s, pods 600 s). No sandbox exec/fs/snapshot API.", "product_class": "paas" }, { "key": "gpu-machine-rtx-4090", "label": "On-demand GPU machine, 1x RTX 4090 (8 vCPU / 64 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 8, "max_ram_gib": 64, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "RTX-4090": 0.44 }, "note": "'From' price $0.44/h per machine incl. 8 vCPU / 64 GiB / 500 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-rtx-a6000", "label": "On-demand GPU machine, 1x A6000 (8 vCPU / 64 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 8, "max_ram_gib": 64, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "RTX-A6000": 0.54 }, "note": "'From' price $0.54/h per machine incl. 8 vCPU / 64 GiB / 500 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-rtx-5090", "label": "On-demand GPU machine, 1x RTX 5090 (12 vCPU / 64 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 12, "max_ram_gib": 64, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "RTX-5090": 0.72 }, "note": "'From' price $0.72/h per machine incl. 12 vCPU / 64 GiB / 500 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-l40s", "label": "On-demand GPU machine, 1x L40S (12 vCPU / 96 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 12, "max_ram_gib": 96, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "L40S": 0.76 }, "note": "'From' price $0.76/h per machine incl. 12 vCPU / 96 GiB / 1000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-rtx-pro-6000", "label": "On-demand GPU machine, 1x RTX PRO 6000 (16 vCPU / 128 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 16, "max_ram_gib": 128, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "RTX-PRO-6000": 1.09 }, "note": "'From' price $1.09/h per machine incl. 16 vCPU / 128 GiB / 1000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-a100-80gb", "label": "On-demand GPU machine, 1x A100 80GB (16 vCPU / 128 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 16, "max_ram_gib": 128, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "A100-80GB": 1.36 }, "note": "'From' price $1.36/h per machine incl. 16 vCPU / 128 GiB / 1000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-h100", "label": "On-demand GPU machine, 1x H100 PCIe (26 vCPU / 200 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 26, "max_ram_gib": 200, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100-PCIe": 1.83 }, "note": "'From' price $1.83/h per machine incl. 26 vCPU / 200 GiB / 1000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ. Pricing page names it H100 PCIE (2026-09-28).", "product_class": "gpu-instance" }, { "key": "gpu-machine-h200", "label": "On-demand GPU machine, 1x H200 (24 vCPU / 192 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 24, "max_ram_gib": 192, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 2.09 }, "note": "'From' price $2.09/h per machine incl. 24 vCPU / 192 GiB / 2000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "gpu-machine-b200", "label": "On-demand GPU machine, 1x B200 (28 vCPU / 256 GiB bundled)", "flags": [], "os": [ "linux" ], "regions": [ "us" ], "gpu_only": true, "plan_limits_exempt": true, "features": { "isolation": null, "image_oci": null }, "plans": null, "pricing": "resource", "vcpu_h": 0, "ram_gib_h": 0, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": 28, "max_ram_gib": 256, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": null, "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 4.11 }, "note": "'From' price $4.11/h per machine incl. 28 vCPU / 256 GiB / 2000 GB NVMe (https://www.beam.cloud/pricing, verified 2026-09-28). Single-GPU machine; price is the whole machine (CPU/RAM included, so vcpu_h=ram_gib_h=0). Docs GPU page lists different on-demand prices (A100-80 $1.49, H100 $3.63, 4090 $0.66). Not a sandbox; capacity may differ.", "product_class": "gpu-instance" }, { "key": "byoc", "label": "BYOC management fee (compute billed by your cloud)", "flags": [ "sales", "alt" ], "os": [ "linux" ], "regions": null, "gpu_only": false, "features": { "isolation": "container" }, "pricing": "resource", "vcpu_h": 0.019, "ram_gib_h": 0.009, "cpu_basis": "alloc", "ram_basis": "alloc", "active_floor": 0, "ram_per_vcpu": [ null, null ], "ram_fixed_per_vcpu": null, "vcpu_options": null, "min_vcpu": null, "max_vcpu": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": false, "always_on_month_per_instance": null, "commit_note": "Book a call; your AWS/GCP/Azure instance cost is NOT included.", "min_billed_seconds": null, "granularity_s": null, "start_fee": 0, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": "$0.019/hr per vCPU (per vCPU, not per core) + $0.009/hr per GB RAM. Example g5.2xlarge 8 vCPU/32 GB = $0.44/h fee. Understates total cost: cloud VM bill is extra. Fee only: + your cloud bill (AWS/GCP/Azure instance cost not included; e.g. the underlying g5.2xlarge is billed by AWS separately).", "product_class": "byoc" } ], "storage": { "included_disk_gib": null, "disk_gib_month": null, "disk_billed_when_stopped": false, "snapshot_gib_month": 0, "paused_ram_billed": false }, "network": { "egress_free_gib": null, "egress_gib": 0, "ipv4_month": null, "ipv4_available": false, "internet": { "default": "configurable", "evidence": [ { "quote": "block all outbound network access from your Sandbox while still allowing inbound connections to exposed ports", "url": "https://docs.beam.cloud/v2/sandbox/networking" }, { "quote": "specify an allow list of CIDR ranges that your Sandbox is permitted to connect to", "url": "https://docs.beam.cloud/v2/sandbox/networking" }, { "quote": "There are no egress or bandwidth fees on any plan.", "url": "https://www.beam.cloud/pricing" } ] } }, "plans": [ { "name": "Developer", "fee": 0, "flags": [], "fee_is_credit": false, "included_usd": 0, "per_seat": 0, "concurrency": 30, "max_session_h": null, "max_vcpu": null, "max_ram_gib": null, "sizes_allowed": null, "note": "Pay-as-you-go (payment method required). 30 CPU / 5 GPU concurrent containers; 1 seat; 30-day logs. A $30/month free credit was advertised through 2026-08-29 but is absent from the pricing page by 2026-09-12 and on 2026-09-28 (not deducted)." }, { "name": "Team", "fee": 89, "flags": [], "fee_is_credit": false, "included_usd": 0, "per_seat": 25, "concurrency": 1000, "max_session_h": null, "max_vcpu": null, "max_ram_gib": null, "sizes_allowed": null, "note": "$89/month plus usage (fee is not credit). 1,000 CPU / 50 GPU concurrent containers; 3 seats included, $25 per extra seat; live chat support." }, { "name": "Growth", "fee": null, "flags": [ "sales" ], "fee_is_credit": null, "included_usd": null, "per_seat": 0, "concurrency": null, "max_session_h": null, "max_vcpu": null, "max_ram_gib": null, "sizes_allowed": null, "note": "Contact us: unlimited CPU containers, 1,000+ GPU, unlimited seats, 1-year log retention, private Slack. Volume discounts above $10K/month spend (unpublished)." } ], "extras": [ { "item": "Extra seat (Team)", "price": "$25/seat/month" }, { "item": "Storage volumes above 1 TB", "price": "$0.021/GB-month" }, { "item": "Reserved GPU clusters (8-GPU nodes, InfiniBand)", "price": "contact sales" } ], "free": { "monthly_credit": null, "one_time_credit": 0 }, "features": { "linux": true, "windows": false, "macos": false, "arm64": false, "gpu": true, "gpu_types": [ "T4", "A10", "RTX-4090", "RTX-5090", "H100", "H200", "A100-80GB", "L40S", "A6000", "RTX-PRO-6000", "B200" ], "snapshot": "mem", "pause_resume": false, "fork": true, "persistent_disk": true, "volumes": true, "auto_stop_idle": false, "max_session_h": null, "custom_image": true, "docker_inside": true, "nested_virt": false, "root": true, "systemd": false, "browser": false, "desktop": false, "computer_use": false, "code_interpreter": false, "agent_harnesses": false, "ssh": true, "public_ipv4": false, "inbound_https": true, "custom_domain": false, "raw_tcp_inbound": false, "egress_allowlist": true, "static_egress_ip": true, "private_network": null, "self_host": true, "byoc": true, "open_source": true, "regions": [ "us" ], "soc2": true, "hipaa": true, "sso": false, "lang_python": true, "lang_node": true, "lang_go": null, "lang_rust": null, "lang_java": null, "gpu_desktop": false, "credential_injection": null, "egress_http_rules": false, "wake_on_request": false, "live_resize": false, "fork_running": true, "memory_snapshot_fork": true, "webhooks": true, "scheduled_wakeups": true, "audit_logs": null, "eu_data_residency": null, "zero_data_retention": null, "scoped_api_keys": null, "spend_limits": null, "mcp_server": true, "isolation": "gvisor", "image_oci": true, "snapshot_auto": false, "snapshot_on_demand": true, "harness_api": false, "egress_allowlist_level": "ip", "ingress_rules": false, "secret_proxy": false, "volume_attach": true, "volume_shared": true, "anti_bot": false, "captcha_solving": false, "egress_ip_type": "static-datacenter", "own_agent_api": false, "guest_firewall": false, "max_session_unlimited": true, "fast_boot": false, "open_internet": true, "usage_api": false }, "perf": { "cpu_runs_s": null, "cold_start_ms": null, "burst_tti_ms": null, "dax_s": null }, "perf_source": "Several runtimes (sandbox, GPU sandbox, serverless containers, GPU machines): perf is on the CPU sandbox mode only.", "knobs": [ { "id": "keep_warm_tail", "label": "keep_warm idle tail (x session hours)", "kind": "slider", "min": 1, "max": 4, "step": 0.05, "default": 1, "unit": "x", "options": null, "effect": "hours_factor", "modes": [ "on-demand", "sandbox-gpu" ], "why": "Beam bills a sandbox until terminate() or until its keep_warm_seconds TTL expires. The SDK default TTL is 600 s (10 min) when keep_warm_seconds is not set, so a sandbox that is never terminate()d bills a 10-minute idle tail; -1 never expires; docs examples use 30 min-2 h. Factor = (session + TTL) / session; 1 = explicit terminate().", "source": "https://docs.beam.cloud/v2/sandbox/configuration" } ], "caveats": [ "CPU is priced per physical core = 2 vCPU; vcpu_h is the per-core price / 2. The SDK documents cpu= as \"The number of CPU cores allocated to the container. Default is 1.0\" (default memory 128 MiB), so cpu=1 = 1 billed core = 2 vCPU as priced. Whether a 'core' in the SDK is a full physical core (2 hyperthreads) is not stated explicitly.", "Dated price change: Beam blogs of June 2026 quote sandbox CPU $0.0000528/core-s ($0.19/core-h) and RAM $0.0000056/GB-s; current page $0.0000375 and $0.0000064 (CPU -29%, RAM +14%).", "Sandbox CPU/RAM are 3x the serverless container rate (alt mode). The page's cost calculator uses the serverless rate, not the sandbox rate.", "free.monthly_credit set null (2026-09-28): the $30/month Developer credit appears on the 2026-08-29 Wayback capture of beam.cloud/pricing but not on 2026-09-12 or the live page; legacy-account entitlement unknown, so no credit is deducted. https://web.archive.org/web/20260829214155/https://www.beam.cloud/pricing , https://web.archive.org/web/20260912172148/https://www.beam.cloud/pricing", "Idle keep_warm TTL tail is billed at full rate until expiry (sandbox examples 30 min-2 h; -1 never expires).", "Snapshots (filesystem images and memory snapshots) included free; volumes free up to 1 TB then $0.021/GB-month; egress free.", "Sandbox GPU pricing is not published; the sandbox-gpu regime is left unpriced (gpu rates null). On-demand GPU machine prices conflict between pricing page and docs GPU page (pricing page used).", "Isolation: default sandboxes are containers (security page 2026-10-14: \"Workloads run in isolated containers.\"; Beam's blog says gVisor; CLI prompt shows runc). The pricing FAQ still says \"non-root containers\" but the security page now says \"a custom image is not guaranteed to run as a non-root user\"; beam shell gives a root prompt, so root=true. No systemd in sandboxes (Beam VMs have systemd).", "Max sandbox size and max lifetime unpublished. Workloads run in the US unless you choose a region \"where the platform offers one\"; no non-US region is listed anywhere, so every mode is US-only and eu_data_residency is unclear. No region multiplier published.", "BYOC fee is per vCPU and excludes the cloud VM bill. Self-hosted beta9 (AGPL) has no Beam fee.", "perf from benchmarks.json (computesdk cold start 105 ms, burst 1,324 ms, dax 64.9 s); no hpc CPU result.", "unverified: may bill active CPU/RAM. Beam's sandbox docs say \"You only pay for what you use. Start small and scale up as needed.\" but no page says whether billing uses requested or consumed cpu/memory; alloc (requested) assumed as the upper bound.", "BYOC regime is a management fee only: add your own cloud bill (compute, storage, egress from AWS/GCP/Azure).", "Sandbox root-filesystem size and price are not published (pricing/billing/sandbox-configuration pages re-checked 2026-09-28); only storage volumes are priced (free up to 1 TB, then $0.021/GB-month). Disk left unknown.", "Durable disks: node-local SSD persistent directory snapshotted to object storage, single writer, attachable to Pod, function, endpoint, task_queue and Sandbox; price not published.", "Sandbox(use_vm=True) runs a sandbox in a microVM instead of a container (\"Run the sandbox in its own virtual machine instead of a container. Only CPU sandboxes are supported;\", https://raw.githubusercontent.com/beam-cloud/beta9/main/sdk/src/beta9/abstractions/sandbox.py; docs VM overview: \"Container runtime, or microVM with `use_vm=True`\"). No separate rate is published for it, so it is assumed to bill at the sandbox rate and is not a separate mode; card isolation stays gvisor (the default).", "Beam VMs (vm mode, added 2026-10-14) are a separate product from sandboxes: per-vCPU $0.004275/h + $0.0021375/GiB-h with a 50 GiB root disk included (pricing page). Whether a stopped VM's root disk or its snapshots are ever billed is not stated (storage is \"Included\" up to 1 TB); card-level storage fields stay as they are. VM concurrency limits are not published separately from plan container limits." ], "sources": [ "https://www.beam.cloud/pricing", "https://docs.beam.cloud/v2/resources/pricing-and-billing", "https://docs.beam.cloud/v2/resources/faq", "https://docs.beam.cloud/v2/sandbox/configuration", "https://docs.beam.cloud/v2/environment/gpu", "https://docs.beam.cloud/v2/self-hosting/overview", "https://www.beam.cloud/blog/2026-sandbox-guide", "https://www.beam.cloud/blog/modal-pricing-explained", "https://docs.beam.cloud/v2/sandbox/networking.md", "https://docs.beam.cloud/v2/security/security.md", "https://docs.beam.cloud/v2/vm/overview", "https://docs.beam.cloud/v2/vm/cli" ], "fleet": { "default_max_instances": 30, "default_max_vcpu": null, "default_max_gpus": null, "raise": "self-serve", "raise_time_days": null, "verification": null, "tiers": "Developer: 30 CPU containers and 5 GPU containers; Team ($89/month): 1,000 CPU containers and 50 GPU containers; Growth: unlimited CPU and 1,000+ GPU containers via contact.", "notes": "default_max_instances refers to CPU-container concurrency. GPU-container allowance is not a GPU-unit quota, so default_max_gpus remains null. Team publishes a route to 50/300/1,000 CPU containers, subject to funding/capacity. No quota-increase turnaround stated.", "sources": [ "https://www.beam.cloud/pricing", "https://docs.beam.cloud/v2/resources/faq" ], "researched": "2026-09-28" }, "audit_2026_09_29": { "applied": 1, "source": "checked against the provider's live pricing and docs pages", "reviewed": 7 }, "feature_evidence": { "snapshot_auto": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "# Auto-terminate after 1 hour\nsandbox = Sandbox(keep_warm_seconds=3600)", "note": "Sandboxes terminate when keep_warm_seconds expires; no pause/sleep with state capture is documented, snapshots are explicit SDK calls. Only exception: attached DurableDisks are snapshotted to object storage on container stop ('Beam snapshots a disk to object storage each time the container using it stops', https://docs.beam.cloud/v2/data/durable-disks.md). That is disk-only persistence of an extra attached disk, not the sandbox's state, so not counted here.", "checked": "2026-09-29" }, "snapshot_on_demand": { "value": true, "source": "https://docs.beam.cloud/v2/sandbox/snapshots.md", "quote": "You can also create a memory snapshot of a running Sandbox, which will capture the state of the sandbox's memory - including all running processes and exposed ports.", "note": "Two kinds: create_image_from_filesystem() (filesystem -> image id usable as base image for any Sandbox/app) and snapshot_memory() (memory + processes + exposed ports; SDK docstring says also GPU state). New sandboxes boot from them via Image.from_id / Sandbox().create_from_memory_snapshot(), i.e. fork. Snapshots are free per pricing page.", "checked": "2026-09-29" }, "harness_api": { "value": false, "source": "https://raw.githubusercontent.com/beam-cloud/beta9/main/sdk/src/beta9/abstractions/sandbox.py", "quote": "A sandboxed container for running Python code or arbitrary processes.", "note": "The Sandbox SDK exposes only process exec/run_code, filesystem, ports, snapshots, network, Docker. No API to start/stream/resume Claude Code/Codex/etc. Beam's 'agents' docs are a petri-net agent framework, and the new MCP server (beta9 PR #1920/#1942) lets an outside agent drive Beam (deploy, pods, secrets), not a harness running inside a sandbox. 'Managed agents' in commits are Beam's worker-machine agents.", "checked": "2026-09-29" }, "egress_allowlist": { "value": true, "source": "https://docs.beam.cloud/v2/sandbox/networking.md", "quote": "For more fine-grained control, you can specify an allow list of CIDR ranges that your Sandbox is permitted to connect to. All other outbound traffic will be blocked.", "note": "block_network=True (no outbound) or allow_list of max 10 CIDRs (IPv4/IPv6), no hostnames. Updatable at runtime via update_network_permissions() without restart. Same for Pods.", "checked": "2026-09-29", "level": "ip" }, "egress_http_rules": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/networking.md", "quote": "Must use proper CIDR notation (e.g., \"8.8.8.8/32\" for a single IP, \"10.0.0.0/8\" for a range)", "note": "Only CIDR-level egress controls are documented; nothing at HTTP method/path/header level.", "checked": "2026-09-29" }, "ingress_rules": { "value": false, "corrected_from": true, "source": "https://docs.beam.cloud/v2/sandbox/networking.md", "quote": "Exposed sandbox URLs are public. Add authentication inside the application before exposing sensitive data or operations.", "note": "Sandboxes: you only choose which ports to expose (ports=[...] or expose_port()); the docs now state the URLs are public and that outbound restrictions do not authenticate inbound requests. The SDK still has an undocumented `authorized` flag (\"Whether the sandbox should be authorized for external access. Default is False.\", https://raw.githubusercontent.com/beam-cloud/beta9/main/sdk/src/beta9/abstractions/sandbox.py) but the docs give it no auth effect, so it is not counted. No IP allowlist for inbound. Beam VMs (vm mode) do have token-protected ports (`--protected-port PORT`: \"Publish a token-protected port\", https://docs.beam.cloud/v2/vm/cli), set on that mode.", "checked": "2026-10-14" }, "secret_proxy": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "You can attach secrets to your sandbox using Beam's secret management system - they will be exposed as environment variables inside the Sandbox", "note": "Secrets are only env vars inside the container; no credential-injecting egress proxy documented.", "checked": "2026-09-29" }, "volume_attach": { "value": true, "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "Mount fast storage volumes to persist files between sessions", "note": "Three kinds attachable to a Sandbox: distributed Volumes, CloudBucket (your S3-compatible bucket), DurableDisk (node-local SSD snapshotted to object storage, 'The `disks` parameter is accepted by ... Sandbox'). Volumes free to 1 TB then $0.021/GB-month.", "checked": "2026-09-29" }, "volume_shared": { "value": true, "source": "https://docs.beam.cloud/v2/data/durable-disks.md", "quote": "If several containers need to read and write the same files at the same time, use a [Volume](/v2/data/volume) instead. Volumes are shared, distributed storage.", "note": "Volumes are read-write shared across containers/sandboxes (writes visible to others within ~60 s); CloudBuckets also shared. DurableDisks: one writer at a time, any number of read-only mounts.", "checked": "2026-09-29" }, "nested_virt": { "value": false, "source": "https://docs.beam.cloud/v2/security/security.md", "quote": "Workloads run in isolated containers. The effective user depends on the image and runtime configuration; a custom image is not guaranteed to run as a non-root user.", "note": "Default sandboxes are gVisor/runc containers (card isolation: gvisor). No /dev/kvm or nested-virtualization capability is documented anywhere in the SDK/docs; a container sandbox does not expose hardware virtualization to the guest. The opt-in microVM runtimes (Sandbox(use_vm=True), Beam VMs) document no nested KVM either. not offered.", "checked": "2026-10-14" }, "anti_bot": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/overview.md", "quote": "Sandboxes are ultra-fast, Python-native environments for running any workload - with GPUs, networking, and persistent storage - in seconds.", "note": "Beam is a serverless compute/sandbox platform for running code, GPUs and containers; it has no browser (browser=false) or managed-browser product, so stealth/fingerprinting-evasion against bot detection is not offered. not offered: outside the product's scope.", "checked": "2026-09-30" }, "captcha_solving": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/overview.md", "quote": "Sandboxes are ultra-fast, Python-native environments for running any workload - with GPUs, networking, and persistent storage - in seconds.", "note": "No browser/computer-use product exists on Beam (browser=false, desktop=false); built-in CAPTCHA solving is not part of a general-purpose compute/sandbox platform. not offered: outside the product's scope.", "checked": "2026-09-30" }, "egress_ip_type": { "value": "static-datacenter", "source": "https://docs.beam.cloud/v2/pod/networking.md", "quote": "Pods are served in a static IP range, making it possible to whitelist the Beam IP range from the client.", "note": "Outbound traffic comes from AWS/GCP datacenter ranges (security page: infrastructure on AWS/GCP). Documented explicitly as a fixed/static IP range for Pods (contact Beam for the range); Sandbox docs don't repeat this, but sandboxes share the same network egress path, so a fixed datacenter range is the best-supported answer rather than plain rotating datacenter IPs.", "checked": "2026-09-30" }, "fork_running": { "value": true, "source": "https://docs.beam.cloud/v2/sandbox/snapshots.md", "quote": "You can also create a memory snapshot of a running Sandbox, which will capture the state of the sandbox's memory - including all running processes and exposed ports.", "note": "The docs example calls sandbox.snapshot_memory() on a sandbox that is still running (a server process is executing and a port is exposed) and only calls terminate() afterward; the snapshot is not preceded by stopping the sandbox. The resulting snapshot can be used via create_from_memory_snapshot() to clone a new sandbox with processes/state intact, i.e. forking a running machine without stopping it.", "checked": "2026-09-30" }, "wake_on_request": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "Manual termination only\nsandbox = Sandbox(keep_warm_seconds=-1)", "note": "For the Sandbox product (this card's primary mode), a sandbox that hits its keep_warm_seconds TTL (or is terminate()d) is gone; there is no documented mechanism for an external request to automatically wake/recreate a terminated sandbox with its prior state (memory snapshots must be explicitly restored via create_from_memory_snapshot()). This matches the card's own mode-level features (auto_stop_idle sandbox modes: wake_on_request=false). Beam's separate serverless containers (Endpoints/pods, an alt/paas mode) do scale from zero on an incoming request (wake_on_request=true there), but that is a different product surface than the Sandbox.", "checked": "2026-09-30" }, "live_resize": { "value": false, "source": "https://docs.beam.cloud/v2/reference/py-sdk.md", "quote": "SandboxInstance.update_ttl()", "note": "The SandboxInstance API only exposes expose_port(), list_urls(), update_network_permissions(), update_ttl(), terminate(), create_image_from_filesystem(), snapshot_memory() -- no update_cpu/update_memory/resize method. The configuration docs tell users to create a new, bigger sandbox instead of resizing ('If you need more power, create a new sandbox'), confirming CPU/RAM cannot be changed on a running sandbox.", "checked": "2026-09-30" }, "own_agent_api": { "value": false, "source": "https://docs.beam.cloud/v2/agents/introduction.md", "quote": "Beam is launching a new type of agent framework that is stateful and has concurrency built-in.", "note": "Beam's 'agents' feature is a petri-net orchestration framework (Bot/BotLocation/BotContext) for building YOUR OWN agents on top of Beam compute -- it is not a first-party API that drives a Beam-owned browser/computer-use/research agent. Beam has no browser or desktop product. not offered: outside the product's scope.", "checked": "2026-09-30" }, "lang_node": { "value": true, "source": "https://docs.beam.cloud/v2/sandbox/overview.md", "quote": "Running a Node.js server\nYou can run arbitrary code on Beam. It doesn't need to be Python!\n... sb = Sandbox(image=Image().from_registry(\"node:20\")).create()", "note": "Node.js is a documented first-class quickstart runtime for Sandboxes (via image), and Beam additionally ships a first-party TypeScript SDK (@beamcloud/beam-js) to drive Sandboxes/Pods/etc. from Node.js code.", "checked": "2026-09-30" }, "guest_firewall": { "value": false, "source": "https://docs.beam.cloud/v2/security/security.md", "quote": "Workloads run in isolated containers. The effective user depends on the image and runtime configuration; a custom image is not guaranteed to run as a non-root user.", "note": "derived: the default sandbox is a gVisor container, not a full VM, so nftables/iptables inside the guest are not expected to work reliably even as root (root=true per the CLI shell transcript); outbound policy is enforced provider-side via allow_list/block_network (egress_allowlist). No page documents an in-guest firewall.", "checked": "2026-10-14" }, "static_egress_ip": { "value": true, "source": "https://docs.beam.cloud/v2/pod/networking.md", "quote": "Pods are served in a static IP range, making it possible to whitelist the Beam IP range from the client.", "note": "Documented for Pods only ('For the static IP range, send us a message in Slack'): a fixed/static range rather than a dedicated single IP per machine, and Sandbox docs do not repeat this claim explicitly, but it indicates Beam's egress infrastructure uses a stable, whitelistable IP range rather than fully arbitrary rotating IPs.", "checked": "2026-09-30" }, "eu_data_residency": { "value": null, "was": true, "source": "https://docs.beam.cloud/v2/security/security.md", "quote": "Customer workloads run in the United States unless you choose a specific region or compute pool where the platform offers one.", "note": "doubtful \"yes\" could not be confirmed: the page allows a non-US region \"where the platform offers one\" but names no EU region, and no other docs/pricing page lists one. Self-hosting (beta9) or BYOC in your own EU cloud account would keep data in the EU, but that is not hosted EU residency. Left unclear.", "checked": "2026-10-14" }, "root": { "value": true, "corrected_from": false, "source": "https://docs.beam.cloud/v2/reference/cli.md", "quote": "$ beam shell app.py:handler ... Welcome to Ubuntu 22.04.5 LTS (GNU/Linux 6.8.0-51-generic x86_64) root@runc:/mnt/code#", "note": "Beam's own CLI reference shows `beam shell` dropping you into a root@ prompt inside the container. The security page no longer promises a non-root user (\"a custom image is not guaranteed to run as a non-root user\", https://docs.beam.cloud/v2/security/security.md, 2026-10-14); the pricing FAQ still says \"run in non-root containers\". The CLI transcript is the concrete behavior, so root stays true.", "checked": "2026-09-30" }, "pause_resume": { "value": false, "source": "https://docs.beam.cloud/v2/reference/py-sdk.md", "quote": "SandboxInstance.terminate() Terminate the container associated with this sandbox instance.", "note": "The SDK exposes only create()/terminate() (plus a keep_warm_seconds TTL and update_ttl()); there is no stop/pause/suspend call that freezes a running sandbox and later resumes it at reduced/no cost. The configuration docs' phrase 'pause and resume work' is just marketing for setting keep_warm_seconds=-1 (manual termination), not a real pause API - confirms no.", "checked": "2026-09-30" }, "desktop": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/introduction.md", "quote": "Beam lets you run functions, REST APIs, task queues, and sandboxes on CPUs and GPUs, end to end.", "note": "Not offered: outside the product's scope. Beam is a code-execution/serverless-compute sandbox (process exec, filesystem, ports, GPUs); no VNC/RDP/graphical desktop is mentioned anywhere in the docs.", "checked": "2026-09-30" }, "browser": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/introduction.md", "quote": "Beam lets you run functions, REST APIs, task queues, and sandboxes on CPUs and GPUs, end to end.", "note": "Not offered: outside the product's scope. No managed/real browser is bundled or offered; the docs' web-scraping example just installs whatever scraping library the user chooses inside a plain Linux container.", "checked": "2026-09-30" }, "computer_use": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/introduction.md", "quote": "Beam lets you run functions, REST APIs, task queues, and sandboxes on CPUs and GPUs, end to end.", "note": "Not offered: outside the product's scope. No screenshot+click desktop-driving API exists; Beam's 'agents' page is a petri-net task-orchestration framework, not a computer-use API.", "checked": "2026-09-30" }, "public_ipv4": { "value": false, "source": "https://docs.beam.cloud/v2/sandbox/networking.md", "quote": "url = sb.expose_port(8000) ... The URL will be something like: https://384ced3c-f837-4429-bada-39e0b965c9f4-8000.app.beam.cloud", "note": "Sandboxes are reached only via SSL-terminated *.app.beam.cloud hostnames, not an assigned public IPv4 address. (Pods separately sit behind a shared static outbound IP range on request, but that's egress, not a per-machine public IPv4 - docs.beam.cloud/v2/pod/networking.md: 'Pods are served in a static IP range'.) Confirms no.", "checked": "2026-09-30" }, "agent_harnesses": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/add-to-cursor-claude.md", "quote": "Claude Code", "note": "The only Claude/Cursor integration is adding Beam's own documentation as an MCP tool inside a local Claude Code/Cursor session on the user's machine - not a preinstalled coding-agent template running inside a Beam sandbox. No page offers Claude Code/Codex/OpenCode as a sandbox template.", "checked": "2026-09-30" }, "windows": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/installation.md", "quote": "You can install Beam on Windows using Windows Subsystem for Linux (WSL).", "note": "The only 'Windows' mention is running the local Beam CLI on a developer's Windows machine via WSL; the remote sandbox/container itself is Linux only (Ubuntu 22.04, per the CLI shell transcript). Not offered: outside the product's scope.", "checked": "2026-09-30" }, "macos": { "value": false, "source": "https://docs.beam.cloud/v2/getting-started/introduction.md", "quote": "Beam lets you run functions, REST APIs, task queues, and sandboxes on CPUs and GPUs, end to end.", "note": "Not offered: outside the product's scope. Beam runs customer workloads on AWS/GCP Linux infrastructure (docs.beam.cloud/v2/security/security.md: 'Beam runs on infrastructure provided by Amazon Web Services and Google Cloud'); no macOS machine type exists.", "checked": "2026-09-30" }, "arm64": { "value": false, "source": "https://docs.beam.cloud/v2/environment/custom-images.md", "quote": "base_image=\"docker.io/pytorch/pytorch:2.2.1-cuda12.1-cudnn8-devel\"", "note": "No architecture flag or ARM/aarch64/Graviton option appears anywhere in the docs (custom-images, custom-registries, resources, py-sdk); base images and GPU examples are all standard x86_64 (CUDA) images, and the CLI shell transcript shows 'x86_64'. Treated as not offered.", "checked": "2026-09-30" }, "sso": { "value": false, "source": "https://docs.beam.cloud/v2/security/security.md", "quote": "All members of a workspace currently have the same permissions, and single sign-on is not yet available; remove members and rotate tokens when someone leaves your team.", "note": "Explicit first-party statement that SSO is not yet available. Confirms no.", "checked": "2026-09-30" }, "max_session_h": { "value": "none", "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "# Manual termination only\nsandbox = Sandbox(keep_warm_seconds=-1)", "note": "No published hard cap on sandbox lifetime. keep_warm_seconds=-1 lets a sandbox run until you explicitly terminate() it ('Use manual termination for: Long-running training jobs, Collaborative development sessions, When you need to pause and resume work'). Docs/pricing/FAQ pages re-checked 2026-09-30 have no mention of a maximum session duration or forced eviction time, consistent with the card's own caveat 'Max sandbox size and max lifetime unpublished.' Treated as no limit rather than null since the platform explicitly supports indefinite manual-only termination.", "checked": "2026-09-30" }, "fast_boot": { "value": false, "source": "https://www.computesdk.com/benchmarks", "quote": "burst of 100: p95 2944 ms, 100% started", "note": "ComputeSDK sandbox benchmark 2026-09-25", "checked": "2026-09-25" }, "usage_api": { "value": false, "source": "https://docs.beam.cloud/v2/reference/platform-api", "quote": "Metrics endpoints (GET /metrics/{ws}/stub-timeseries, workspace-timeseries, pool-timeseries) return resource time series (CPU, memory, GPU, network) but not billable usage units or costs. Billing is visible in the dashboard only; no programmatic billing/cost API is documented.", "note": "granularity: none; cost: no; Beam exposes metrics for monitoring (CPU/memory/GPU/network per container) via REST API and logs/task status endpoints, but no dedicated billing usage API that returns billable quantities (CPU-hours, RAM-hours, GPU-hours) or cost per sandbox/tag/project. Usage is only accessible through the account dashboard.", "confidence": "high", "checked": "2026-10-01" }, "raw_tcp_inbound": { "value": false, "corrected_from": true, "source": "https://docs.beam.cloud/v2/sandbox/networking.md", "quote": "You can run web services and expose them to the internet behind SSL-terminated endpoints.", "note": "Exposed ports are reached only as https://-.app.beam.cloud URLs; no raw TCP/UDP port forwarding or public IP is documented for sandboxes (the on-demand mode already had raw_tcp_inbound=false). Beam VMs offer private TCP via vm.tunnel()/portForward() over the Beam CLI/SDK, which is not public inbound TCP.", "checked": "2026-10-14" }, "auto_stop_idle": { "value": false, "corrected_from": true, "source": "https://docs.beam.cloud/v2/sandbox/configuration.md", "quote": "`keep_warm_seconds` is the sandbox's TTL, not an idle timeout that every command automatically resets.", "note": "Matches the sandbox modes' auto_stop_idle=false: a sandbox runs (and bills) until terminate() or TTL expiry whether idle or not. Beam VMs (vm mode) do have an idle sleep timeout (`--ttl SECONDS`: \"Idle sleep timeout; unset or `0` keeps running.\"), set true on that mode only.", "checked": "2026-10-14" }, "regions": { "value": [ "us" ], "source": "https://docs.beam.cloud/v2/security/security.md", "quote": "Customer workloads run in the United States unless you choose a specific region or compute pool where the platform offers one.", "note": "No list of non-US regions is published on the security, pricing, sandbox or VM pages (2026-10-14), so modes previously listing us/eu/asia were cut to us to match features.regions and locations.", "checked": "2026-10-14" } }, "locations": [ { "country": "US", "state": null, "city": null, "provider_region": "us", "self_serve": true, "choosable": true } ], "locations_source": "https://docs.beam.cloud/v2/security/security.md", "locations_quote": "Customer workloads run in the United States unless you choose a specific region or compute pool where the platform offers one.", "locations_checked": "2026-10-14" }