{ "features_resolved": { "snapshot_any": false, "snapshot_mem": false, "fork": false, "pause_resume": false, "persistent_disk": true, "whole_disk": false, "persist_scope": true, "volumes": true, "auto_stop_idle": false, "long_sessions": true, "custom_image": true, "image_snapshot": false, "image_oci": true, "vm_isolation": false, "docker_inside": true, "nested_virt": true, "root": null, "systemd": null, "browser": false, "desktop": false, "computer_use": false, "code_interpreter": null, "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": true, "inbound_https": true, "custom_domain": null, "raw_tcp_inbound": null, "egress_allowlist": false, "open_internet": null, "static_egress_ip": true, "private_network": null, "self_host": false, "byoc": false, "open_source": false, "soc2": true, "hipaa": true, "sso": true, "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": null, "credential_injection": null, "egress_http_rules": false, "wake_on_request": false, "live_resize": false, "fork_running": false, "memory_snapshot_fork": false, "webhooks": null, "scheduled_wakeups": null, "audit_logs": null, "eu_data_residency": true, "zero_data_retention": null, "scoped_api_keys": null, "spend_limits": null, "mcp_server": false, "usage_api": true, "snapshot_auto": false, "snapshot_on_demand": false, "harness_api": false, "own_agent_api": false, "fast_boot": null, "strong_isolation": false, "ingress_rules": true, "guest_firewall": true, "secret_proxy": false, "secret_proxy_any": false, "volume_attach": true, "volume_shared": true, "sidecar": null, "fuse": null, "kernel_features": null, "burst": null }, "company": { "id": "together-gpu", "checked": "2026-09-30", "company": "Together AI, Inc. (Together Computer, Inc.)", "ownership": { "value": "independent", "parent": null, "source": "https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/", "quote": "Together AI — an AI neocloud that rents out Nvidia GPU clusters and other AI-specific infrastructure — has raised an $800 million Series C at an $8.3 billion valuation, the company announced on Wednesday." }, "stage": { "value": "series-c", "source": "https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/", "quote": "Together AI … has raised an $800 million Series C at an $8.3 billion valuation", "confidence": "high", "note": "Last priced round is the July 2026 Series C ($800M, $8.3B valuation)." }, "accelerators": [], "total_funding_usd": { "value": 1207500000, "status": "confirmed", "basis": "Sum of confirmed rounds per TechCrunch: $102.5M Series A (2023) + $305M Series B (~March 2025, $3.3B valuation) + $800M Series C (July 2026, $8.3B valuation).", "source": "https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/", "quote": "Together AI last raised a $305 million Series B at a $3.3 billion valuation about 16 months ago. It actually came out of the gate well-funded, with a $102.5 million Series A funding round led by Kleiner Perkins with participation from Nvidia and Emergence Capital back in 2023." }, "last_round": { "type": "series-c", "amount_usd": 800000000, "date": "2026-07", "leads": [ "Aramco Ventures" ], "status": "confirmed", "source": "https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/", "quote": "The round was led by Aramco Ventures, with participation from Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron, SentinelOne’s S Ventures, and others." }, "investors": [ "Aramco Ventures", "Vista Equity Partners", "General Catalyst", "Emergence Capital", "Nvidia", "March Capital", "Pegatron", "SentinelOne's S Ventures", "Kleiner Perkins", "Lux Capital", "Salesforce Ventures", "SV Angel" ], "portfolio_listings": [ { "source": "Cherry Ventures", "company": "Together", "batch": null, "how": "name" }, { "source": "Emergence Capital", "company": "Together AI", "batch": null, "how": "exact-name" }, { "source": "Kleiner Perkins", "company": "Together.ai", "batch": 2023, "how": "exact-name" }, { "source": "Lux Capital", "company": "Together.xyz", "batch": 2023, "how": "domain" }, { "source": "Salesforce Ventures", "company": "Together AI", "batch": null, "how": "domain" }, { "source": "SV Angel", "company": "Together AI", "batch": null, "how": "domain" } ], "revenue": { "value_usd": 1150000000, "kind": "ARR", "as_of": "2026-Q2", "status": "confirmed", "basis": "Company-reported annualized bookings figure quoted by TechCrunch, not audited revenue.", "source": "https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/", "quote": "The hefty infusion of capital comes as Together AI claims annual bookings of over $1.15 billion as of its last quarter" }, "employees": { "value": null, "source": "" }, "note": "Founded 2022 by Vipul Ved Prakash (CEO, previously sold Topsy to Apple), Percy Liang (Stanford) and Ce Zhang (ETH Zurich/UChicago). Neocloud/GPU-cloud and inference platform for open-source AI models; customers cited include Cursor, Cognition and Decagon." }, "id": "together-gpu", "name": "Together AI", "url": "https://www.together.ai/pricing", "category": "gpu-cloud", "isolation": "container", "modes": [ { "key": "on-demand-clusters-hgx-sxm-0", "label": "H100 GPU cluster, on-demand (8-GPU nodes, per GPU-hour)", "flags": [], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": 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": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100": 3.99 }, "note": "H100 (80 GB) on-demand \"$3.99/hr per GPU\", scale \"8 to 256 GPUs\", self-serve, \"pay hourly, terminate anytime\" (https://www.together.ai/gpu-clusters and https://www.together.ai/pricing, 2026-09-28). Sold as whole bare-metal HGX nodes of 8 GPUs (API: num_gpus 'must be multiple of 8'), so gpu_min_count 8 (a 1-GPU workload pays for 8). Host CPU/RAM come with the node (vcpu_h/ram_gib_h 0); node CPU/RAM spec not published on the pricing page. Billing: 'Hourly billing: Pay per hour of cluster runtime' (docs); rounding of partial hours not stated.", "gpu_min_count": 8, "plan_limits_exempt": true, "plans": [ "Self-serve (GPU clusters, on-demand)" ] }, { "key": "on-demand-clusters-hgx-sxm-1", "label": "H200 HGX/SXM, on-demand (×8 node)", "flags": [], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 5.99 }, "note": "8-GPU node procurement; minimum cluster size/availability verify in cluster console. Source: https://www.together.ai/pricing Added 2026-09-28 (missing-providers audit). 8-GPU instance priced per GPU ($5.99/GPU-h, from research/gpu.json / https://www.together.ai/pricing); gpu_min_count 8 so the engine bills 8 GPUs.", "gpu_min_count": 8 }, { "key": "on-demand-clusters-hgx-sxm-2", "label": "B200 HGX/SXM, on-demand (×8 node)", "flags": [], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 8.19 }, "note": "8-GPU node procurement; minimum cluster size/availability verify in cluster console. Source: https://www.together.ai/pricing Added 2026-09-28 (missing-providers audit). 8-GPU instance priced per GPU ($8.19/GPU-h, from research/gpu.json / https://www.together.ai/pricing); gpu_min_count 8 so the engine bills 8 GPUs.", "gpu_min_count": 8 }, { "key": "on-demand-clusters-hgx-sxm-3", "label": "B300 HGX/SXM, on-demand (×8 node)", "flags": [], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B300": 9.99 }, "note": "8-GPU node procurement; minimum cluster size/availability verify in cluster console. Source: https://www.together.ai/pricing Added 2026-09-28 (missing-providers audit). 8-GPU instance priced per GPU ($9.99/GPU-h, from research/gpu.json / https://www.together.ai/pricing); gpu_min_count 8 so the engine bills 8 GPUs.", "gpu_min_count": 8 }, { "key": "spot-clusters-hgx-sxm-4", "label": "H100 HGX/SXM, spot (×8 node)", "flags": [ "spot", "beta" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100": 1.99 }, "note": "Preemptible (public preview, Kubernetes only). Cannot be bought alone: every cluster needs at least one standard (on-demand or reserved) 8-GPU node, so the cheapest real spot setup is 8 standard GPUs + N x 8 preemptible GPUs; this per-GPU rate understates the true cost unless the workload already needs a standard node. Flat discount vs on-demand (not a bid), metered every 1-2 minutes, drain window of at most 5 minutes, no minimum-lifetime guarantee. Source: https://docs.together.ai/docs/preemptible-compute", "gpu_min_count": 8 }, { "key": "spot-clusters-hgx-sxm-5", "label": "H200 HGX/SXM, spot (×8 node)", "flags": [ "spot", "beta" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 2.99 }, "note": "Preemptible (public preview, Kubernetes only). Cannot be bought alone: every cluster needs at least one standard (on-demand or reserved) 8-GPU node, so the cheapest real spot setup is 8 standard GPUs + N x 8 preemptible GPUs; this per-GPU rate understates the true cost unless the workload already needs a standard node. Flat discount vs on-demand (not a bid), metered every 1-2 minutes, drain window of at most 5 minutes, no minimum-lifetime guarantee. Source: https://docs.together.ai/docs/preemptible-compute", "gpu_min_count": 8 }, { "key": "spot-clusters-hgx-sxm-6", "label": "B200 HGX/SXM, spot (×8 node)", "flags": [ "spot", "beta" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 4.09 }, "note": "Preemptible (public preview, Kubernetes only). Cannot be bought alone: every cluster needs at least one standard (on-demand or reserved) 8-GPU node, so the cheapest real spot setup is 8 standard GPUs + N x 8 preemptible GPUs; this per-GPU rate understates the true cost unless the workload already needs a standard node. Flat discount vs on-demand (not a bid), metered every 1-2 minutes, drain window of at most 5 minutes, no minimum-lifetime guarantee. Source: https://docs.together.ai/docs/preemptible-compute", "gpu_min_count": 8 }, { "key": "spot-clusters-hgx-sxm-7", "label": "B300 HGX/SXM, spot (×8 node)", "flags": [ "spot", "beta" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B300": 4.99 }, "note": "Preemptible (public preview, Kubernetes only). Cannot be bought alone: every cluster needs at least one standard (on-demand or reserved) 8-GPU node, so the cheapest real spot setup is 8 standard GPUs + N x 8 preemptible GPUs; this per-GPU rate understates the true cost unless the workload already needs a standard node. Flat discount vs on-demand (not a bid), metered every 1-2 minutes, drain window of at most 5 minutes, no minimum-lifetime guarantee. Source: https://docs.together.ai/docs/preemptible-compute", "gpu_min_count": 8 }, { "key": "reserved-clusters-8", "label": "H100, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "7-30 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100": 3.69 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-9", "label": "H200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "7-30 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 4.99 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-10", "label": "B200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "7-30 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 7.99 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-11", "label": "H100, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "31-90 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100": 3.45 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-12", "label": "H200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "31-90 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 4.15 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-13", "label": "B200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "31-90 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 7.79 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-14", "label": "H100, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "91-180 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H100": 3.19 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-15", "label": "H200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "91-180 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "H200": 3.99 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "reserved-clusters-16", "label": "B200, reserved (×8 node)", "flags": [ "commit" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "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": null, "max_ram_gib": null, "sizes": null, "pool_tiers": null, "requires_always_on": true, "always_on_month_per_instance": null, "commit_note": "91-180 days", "min_billed_seconds": null, "granularity_s": 3600, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": { "B200": 6.79 }, "note": "Reserved GPU cluster, per GPU-hour, prepaid upfront for the term, non-refundable; sold as 8-GPU HGX nodes (gpu_min_count 8), host CPU/RAM included. Source: https://www.together.ai/pricing (2026-09-29). 181+ days and B300 reserved via sales.", "gpu_min_count": 8 }, { "key": "on-demand-dedicated-inference-17", "label": "H100, on-demand", "flags": [ "alt" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": " Source: https://www.together.ai/pricing ESTIMATOR EXCLUDED: Bundled node count/CPU/RAM constraints are not enforceable in current engine schema. CPU/RAM inclusion not verified. Exact researched rate remains in gpu.json." }, { "key": "on-demand-dedicated-inference-list-18", "label": "H100, on-demand", "flags": [ "alt" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": " Source: https://www.together.ai/pricing ESTIMATOR EXCLUDED: Bundled node count/CPU/RAM constraints are not enforceable in current engine schema. CPU/RAM inclusion not verified. Exact researched rate remains in gpu.json." }, { "key": "on-demand-dedicated-inference-19", "label": "B200, on-demand", "flags": [ "alt" ], "os": [ "linux" ], "regions": null, "gpu_only": true, "features": {}, "pricing": "resource", "vcpu_h": null, "ram_gib_h": null, "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": null, "granularity_s": 60, "start_fee": null, "boot_overhead_s": null, "multiplier": 1, "windows_vcpu_h": null, "gpu": null, "note": " Source: https://www.together.ai/pricing ESTIMATOR EXCLUDED: Bundled node count/CPU/RAM constraints are not enforceable in current engine schema. CPU/RAM inclusion not verified. Exact researched rate remains in gpu.json." } ], "storage": { "included_disk_gib": null, "disk_gib_month": 0.16, "disk_billed_when_stopped": true, "snapshot_gib_month": null, "paused_ram_billed": null }, "network": { "egress_free_gib": null, "egress_gib": 0, "ipv4_month": null, "ipv4_available": null }, "plans": [ { "name": "Self-serve (GPU clusters, on-demand)", "fee": 0, "flags": [], "fee_is_credit": false, "included_usd": 0, "per_seat": 0, "concurrency": null, "max_session_h": null, "max_vcpu": null, "max_ram_gib": null, "sizes_allowed": null, "note": "GPU Clusters on-demand is \"fully self-serve\": \"None - pay hourly, terminate anytime\", up to 256 GPUs, capacity \"based on real-time availability\" (https://www.together.ai/gpu-clusters, 2026-09-28). No platform fee published." } ], "free": { "monthly_credit": 0, "one_time_credit": 0 }, "features": { "linux": true, "windows": false, "macos": false, "arm64": false, "gpu": true, "gpu_types": [ "B200", "B300", "H100", "H200" ], "snapshot": "none", "pause_resume": false, "fork": false, "persistent_disk": true, "volumes": true, "auto_stop_idle": false, "max_session_h": null, "custom_image": true, "docker_inside": true, "nested_virt": true, "root": null, "systemd": null, "browser": false, "desktop": false, "computer_use": false, "code_interpreter": null, "agent_harnesses": false, "ssh": true, "public_ipv4": true, "inbound_https": true, "custom_domain": null, "raw_tcp_inbound": null, "egress_allowlist": false, "static_egress_ip": true, "private_network": null, "self_host": false, "byoc": false, "open_source": false, "regions": [ "us", "eu" ], "soc2": true, "hipaa": true, "sso": true, "lang_python": true, "lang_node": true, "lang_go": null, "lang_rust": null, "lang_java": null, "gpu_desktop": null, "credential_injection": null, "egress_http_rules": false, "wake_on_request": false, "live_resize": false, "fork_running": false, "memory_snapshot_fork": false, "webhooks": null, "scheduled_wakeups": null, "audit_logs": null, "eu_data_residency": true, "zero_data_retention": null, "scoped_api_keys": null, "spend_limits": null, "mcp_server": false, "snapshot_auto": false, "volume_attach": true, "volume_shared": true, "isolation": "container", "image_oci": true, "anti_bot": false, "captcha_solving": false, "egress_ip_type": "datacenter", "snapshot_on_demand": false, "harness_api": false, "own_agent_api": false, "guest_firewall": true, "secret_proxy": false, "max_session_unlimited": true, "ingress_rules": true, "usage_api": true, "persist_scope": "folder", "persist_path": "shared volumes" }, "perf": { "cpu_runs_s": null, "cold_start_ms": null, "burst_tti_ms": null, "dax_s": null }, "knobs": [], "caveats": [ "GPU cluster rates not valid for Code Sandbox. H100 dedicated inference $3.99 is a promotion ending 2026-09-30; $5.49 normal list. Cluster reservation windows are not serverless usage rates.", "Any null count, CPU/RAM, billing increment or minimum is unverified, not unlimited/free.", "This is GPU-product research, not a claim that every feature of the entire vendor documentation was audited. Unverified features are null.", "GPU multi-count bundles, variant/region selection and fractional slices require a SKU-aware estimator; unsupported rows remain unpriced in strict cards.", "Modes with gpu=null are intentionally excluded, not free GPU modes. Never substitute 0 for a null rate. Flags alt indicate DIY/ML infrastructure, not the same thing as managed untrusted-code sandboxes.", "Unknown fee, storage or bandwidth means full monthly total may be unavailable. Rates exclude taxes.", "On-demand (H100/H200/B200/B300) and spot cluster modes are priced per GPU with gpu_min_count 8; reserved modes are priced as always-on commit modes; dedicated-inference modes are unpriced. The gpu-clusters page lists H200/B200 cluster scale as '256 to 1,000 GPUs', so gpu_min_count 8 may understate the real minimum for those.", "Reserved clusters: prepaid upfront, non-refundable, cannot be paused, auto-decommissioned at the end of the term (24 h email); usage beyond the reservation is billed at on-demand rates. Docs say reservations are 1-90 days while the pricing page sells 7-30 / 31-90 / 91-180 days and 181+ via sales, and the marketing page says 'Up to 6 months'.", "The clusters API also accepts RTX_6000_PCI, L40_PCIE and H100_SXM_INF, and GB200/GB300 NVL72 are 'Contact us'; none has a published cluster price, so they are not listed in gpu_types." ], "sources": [ "https://www.together.ai/pricing", "https://www.together.ai/gpu-clusters" ], "fleet": { "default_max_instances": null, "default_max_vcpu": null, "default_max_gpus": null, "raise": "sales", "raise_time_days": null, "verification": null, "tiers": "On-demand self-serve up to 256 GPUs per cluster (H100 '8 to 256 GPUs'; H200/B200 listed '256 to 1,000'), capacity based on real-time availability; reserved self-serve 'Up to 4,000+ GPUs'; 181+ day reservations, GB200/GB300 NVL72 and B300 reserved via sales. The 64-GPU figure came from the March 2025 preview.", "notes": "Per-cluster supported size is not account-wide fresh GPU quota. Current pricing distinguishes on-demand and reserved products; account approval, payment checks and scale-review time not established.", "sources": [ "https://www.together.ai/blog/instant-gpu-clusters", "https://www.together.ai/pricing" ], "researched": "2026-09-28" }, "audit_2026_09_29": { "applied": 6, "source": "checked against the provider's live pricing and docs pages", "reviewed": 40 }, "feature_evidence": { "snapshot_auto": { "value": false, "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Local NVMe disks and node-local storage are ephemeral.", "note": "GPU Clusters (Kubernetes / Slurm-on-Kubernetes): no state capture on stop; node disks are ephemeral and only shared volumes persist. Reserved clusters cannot be paused and are decommissioned at term end.", "checked": "2026-09-29" }, "volume_attach": { "value": true, "source": "https://docs.together.ai/reference/clusters_storages-create.md", "quote": "You can dynamically create and attach volumes to your cluster at cluster creation time, and resize as your data grows.", "note": "Shared volumes sized in whole TiB, resizable, persist beyond cluster deletion and can be reattached to other clusters. Attach at cluster creation (new or existing volume_id).", "checked": "2026-09-29" }, "volume_shared": { "value": true, "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Shared volumes are remote-attached, high-speed filesystems. They are created during cluster setup (or attached from an existing volume) and are accessible from all nodes.", "note": "Read-write; Kubernetes PV with ReadWriteMany (multiple pods across nodes), Slurm mounts it at /home on all compute and login nodes.", "checked": "2026-09-29" }, "isolation": { "value": "container", "source": "https://docs.together.ai/docs/gpu-clusters-management.md", "quote": "Our Kubernetes runtime exposes all GPU devices to all containers on the host", "note": "GPU Clusters are Kubernetes (or Slurm-on-Kubernetes) clusters of dedicated bare-metal nodes; the customer-facing workload unit is a container/pod on those nodes, matching the card's existing top-level isolation value.", "checked": "2026-09-30" }, "image_oci": { "value": true, "source": "https://docs.together.ai/docs/gpu-clusters-quickstart.md", "quote": "kubectl run test --image=ubuntu --command -- sleep infinity", "note": "Any container image can be run as a pod on the cluster.", "checked": "2026-09-30" }, "docker_inside": { "value": true, "source": "https://docs.together.ai/docs/gpu-clusters-management.md", "quote": "Use kubectl to interact with Kubernetes clusters for containerized workloads.", "note": "Workloads are containers/pods on Kubernetes; Slurm compute/login nodes also give sudo-level SSH access to the underlying Linux host where container tooling could be installed.", "checked": "2026-09-30" }, "desktop": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "browser": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "computer_use": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "anti_bot": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "captcha_solving": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "egress_ip_type": { "value": "datacenter", "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "Together AI does not charge for data transfer into or out of your cluster", "note": "Bare-metal/Kubernetes datacenter nodes with InfiniBand + standard datacenter networking; no residential/mobile IP product exists.", "checked": "2026-09-30" }, "snapshot": { "value": "none", "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Local NVMe disks and node-local storage are ephemeral. Data on these drives can be lost at any time during node migrations/recreations, maintenance, or other cluster operations.", "note": "No filesystem or memory snapshot/checkpoint capability for a whole node is documented; only PVC-backed shared volumes persist data.", "checked": "2026-09-30" }, "snapshot_on_demand": { "value": false, "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Local NVMe disks and node-local storage are ephemeral.", "note": "No API/CLI snapshot or checkpoint action for a cluster/node is documented.", "checked": "2026-09-30" }, "fork": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-overview.md", "quote": "Create clusters in minutes with features like real-time scaling, persistent storage, and support for both Kubernetes and Slurm workload managers.", "note": "No cluster/node cloning feature is documented.", "checked": "2026-09-30" }, "fork_running": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-overview.md", "quote": "Create clusters in minutes with features like real-time scaling, persistent storage, and support for both Kubernetes and Slurm workload managers.", "note": "No cluster/node cloning feature is documented.", "checked": "2026-09-30" }, "memory_snapshot_fork": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-overview.md", "quote": "Create clusters in minutes with features like real-time scaling, persistent storage, and support for both Kubernetes and Slurm workload managers.", "note": "No cluster/node cloning feature is documented.", "checked": "2026-09-30" }, "pause_resume": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "Can I pause a cluster to save costs? Reserved clusters cannot be paused. You're charged for the full reservation period. On-demand clusters can be terminated and recreated later, but there's no pause function.", "note": "Explicitly no pause function for either billing type; the involuntary pause on credit exhaustion is not a user-controlled feature.", "checked": "2026-09-30" }, "wake_on_request": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "On-demand clusters: Run until manually terminated. Can be stopped/started anytime. No automatic decommissioning.", "note": "No automatic wake-on-request mechanism is documented; clusters must be started/stopped manually.", "checked": "2026-09-30" }, "persistent_disk": { "value": true, "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Shared volumes (PVC): Persistent. Survives pod restarts, node reboots/migrations/recreations, cluster operations, and even cluster deletion.", "note": "Shared volumes persist; local NVMe/node disk is explicitly ephemeral.", "checked": "2026-09-30" }, "auto_stop_idle": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "On-demand clusters: Run until manually terminated. Can be stopped/started anytime. No automatic decommissioning.", "note": "No idle-based auto-stop is documented; only credit exhaustion or reservation end triggers stop/decommission.", "checked": "2026-09-30" }, "live_resize": { "value": false, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "auto_scale_max_gpus: Maximum number of GPUs to which the cluster can be auto-scaled up.", "note": "Only cluster-level GPU node-count auto-scaling exists; no documented ability to change a single running node's CPU/RAM without recreating it.", "checked": "2026-09-30" }, "agent_harnesses": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope (GPU Clusters is bare-metal/Kubernetes training infra, not a coding-agent sandbox; Together's separate 'Sandbox'/Developer Environments product is not modeled by this card).", "checked": "2026-09-30" }, "harness_api": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "own_agent_api": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope", "checked": "2026-09-30" }, "mcp_server": { "value": false, "source": "https://docs.together.ai/docs/agent-skills.md", "quote": "Make your AI coding agent Together-AI-aware with ready-made skills for code generation and an MCP server for live docs lookup.", "note": "Together's MCP server is for documentation lookup by a coding agent, not a first-party MCP server to control/drive GPU clusters.", "checked": "2026-09-30" }, "lang_python": { "value": true, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "from together import Together\nclient = Together()\nresponse = client.beta.clusters.create(...)", "note": "Together Python SDK manages clusters; Linux cluster nodes run arbitrary Python.", "checked": "2026-09-30" }, "lang_node": { "value": true, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "import Together from \"together-ai\";\nconst client = new Together();\nconst response = await client.beta.clusters.create({...})", "note": "Together TypeScript/JavaScript SDK manages clusters; Linux cluster nodes run arbitrary Node.js.", "checked": "2026-09-30" }, "egress_http_rules": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "Together AI does not charge for data transfer into or out of your cluster", "note": "No HTTP method/path-level egress control is documented for GPU cluster nodes.", "checked": "2026-09-30" }, "guest_firewall": { "value": true, "source": "https://docs.together.ai/docs/gpu-clusters-management.md", "quote": "sudo scontrol update NodeName= State=drain Reason=\"\"", "note": "derived: sudo/root-level SSH access to bare-metal Slurm/Kubernetes worker nodes implies nftables/iptables would work inside.", "checked": "2026-09-30" }, "secret_proxy": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-overview.md", "quote": "Full API access: Manage clusters via REST API or CLI.", "note": "No first-party credential-injection proxy for outbound HTTP calls is documented; credentials are supplied by the user inside the cluster (env vars, files, Kubernetes secrets).", "checked": "2026-09-30" }, "windows": { "value": false, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "nvidia_driver_version, cuda_version ... cluster_type: KUBERNETES, SLURM", "note": "not offered: outside the product's scope; only Linux nodes with NVIDIA drivers/CUDA are offered (card os: linux).", "checked": "2026-09-30" }, "macos": { "value": false, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "nvidia_driver_version, cuda_version ... cluster_type: KUBERNETES, SLURM", "note": "not offered: outside the product's scope; NVIDIA GPU HGX Linux nodes only.", "checked": "2026-09-30" }, "arm64": { "value": false, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "gpu_type: H100_SXM, H200_SXM, RTX_6000_PCI, L40_PCIE, B200_SXM, H100_SXM_INF, B300_SXM", "note": "All published GPU cluster node types are NVIDIA HGX/PCIe x86 servers; no ARM host option is documented.", "checked": "2026-09-30" }, "self_host": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Self-serve AI-ready GPU clusters at scale.", "note": "not offered: outside the product's scope; GPU Clusters is a managed service on Together's own datacenter capacity, not software you self-host.", "checked": "2026-09-30" }, "byoc": { "value": false, "source": "https://www.together.ai/blog/introducing-the-together-enterprise-platform", "quote": "Deploy in any environment, with full control: Deploy on Together Cloud, your Virtual Private Cloud (VPC), or on-premise.", "note": "The Together Enterprise Platform (inference/fine-tuning) supports VPC/on-prem deployment, but GPU Clusters (this card's product) is sold as Together's own bare-metal capacity, not deployed into the customer's own cloud account.", "checked": "2026-09-30" }, "open_source": { "value": false, "source": "https://www.together.ai/gpu-clusters", "quote": "Managed NVIDIA GPU cluster infrastructure for distributed training, fine-tuning, and large-scale inference.", "note": "not offered: outside the product's scope; the GPU Clusters orchestration platform itself is not published under an OSI license (Together does open-source some research code/models, but not this product).", "checked": "2026-09-30" }, "hipaa": { "value": true, "source": "https://www.together.ai/blog/introducing-the-together-enterprise-platform", "quote": "Compliance with major industry standards such as SOC 2, GDPR, and HIPAA", "note": "Stated at the Together Enterprise Platform / company level, not specific to GPU Clusters.", "checked": "2026-09-30" }, "sso": { "value": true, "source": "https://docs.together.ai/docs/sso.md", "quote": "Single Sign-On enables your company to authenticate to your Together Organization through your company's existing Identity Provider (IdP) ... Together supports SSO via SAML and OIDC protocols", "note": "Available for Scale and Enterprise accounts.", "checked": "2026-09-30" }, "eu_data_residency": { "value": true, "source": "https://docs.together.ai/docs/privacy-and-security.md", "quote": "Together supports private networking and VPC-based deployments, including in EU regions.", "note": "Card's existing regions list already includes 'eu'.", "checked": "2026-09-30" }, "max_session_h": { "value": "none", "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "On-demand clusters: Run until manually terminated. Can be stopped/started anytime. No automatic decommissioning.", "note": "Reserved clusters instead have a fixed prepaid term (1-90+ days) after which they are auto-decommissioned, but on-demand clusters (the base mode) have no session time limit.", "checked": "2026-09-30" }, "nested_virt": { "value": true, "source": "https://www.together.ai/gpu-clusters", "quote": "Bare-metal performance, InfiniBand networking, and managed orchestration — with flexible pricing for both on-demand and reserved capacity.", "note": "derived: nodes are dedicated bare-metal HGX servers, and docs give sudo-level SSH access to Kubernetes worker nodes and Slurm compute nodes ('sudo scontrol update NodeName=...'), so /dev/kvm should be usable inside; not explicitly documented as a feature.", "checked": "2026-09-30" }, "public_ipv4": { "value": true, "source": "https://docs.together.ai/reference/clusters-get.md", "quote": "public_ipv4: type: string description: Public IPv4 address of the GPU worker node.", "note": "The cluster API schema returns a public_ipv4 field for both the control-plane node and each GPU worker node (GPUClusterGPUWorkerNode / GPUClusterControlPlaneNode objects).", "checked": "2026-09-30" }, "inbound_https": { "value": true, "source": "https://docs.together.ai/reference/clusters-create.md", "quote": "install_traefik: type: boolean default: false description: Whether to install Traefik ingress controller in the cluster. This field is only applicable for Kubernetes clusters and is false by default.", "note": "Kubernetes clusters can be created with load_balancer set to TRAEFIK, NGINX, or ISTIO (InstanceClusterConfig.load_balancer enum), which are HTTP(S) ingress controllers for routing external traffic to pods/services; disabled (NONE) by default so it is opt-in, not automatic.", "checked": "2026-09-30" }, "ingress_rules": { "value": true, "source": "https://docs.together.ai/docs/gpu-clusters-management.md", "quote": "Key-based commands use a proxy jump host: ssh -J @ssh...cloud.together.ai @", "note": "Node access is mediated through a per-cluster SSH jump host requiring OIDC sign-in or an uploaded SSH key (jumphost_enabled/ssh_ca_enabled in clusters-create.md), plus project-based admin/editor roles gating who can reach cluster resources at all; Kubernetes clusters can also add Traefik/Nginx/Istio as an ingress layer in front of exposed HTTP(S) ports.", "checked": "2026-09-30" }, "egress_allowlist": { "value": false, "source": "https://docs.together.ai/docs/gpu-clusters-billing.md", "quote": "Together AI does not charge for data transfer into or out of your cluster, so you can move datasets, checkpoints, and results freely.", "note": "No provider-side outbound allow/deny-list, no-internet mode, or security-group egress rule feature is documented anywhere in the cluster management, API, or billing docs; free unrestricted data transfer is described instead.", "checked": "2026-09-30" }, "static_egress_ip": { "value": true, "source": "https://docs.together.ai/reference/clusters-get.md", "quote": "public_ipv4: type: string description: Public IPv4 address of the GPU worker node.", "note": "derived: each bare-metal node holds its own dedicated public_ipv4 (not a shared NAT gateway pool), so outbound traffic from that node uses the same stable address for the node's lifetime; not explicitly documented as a persistent/guaranteed egress IP feature, and node recreation could change it.", "checked": "2026-09-30" }, "usage_api": { "value": true, "source": "https://docs.together.ai/reference/billing-usage.md", "quote": "Returns an organization's billing usage for a month as cost-annotated line items grouped into time windows. Resource identifiers for attribution, varying by product. Product_name and cost in USD returned for each line item.", "note": "granularity: per-sandbox; cost: yes; GET /billing/usage REST API is documented and returns billing usage with cost-annotated line items that include resource attributes. API is in beta, requires per-org enablement. Documentation explicitly shows attributes vary by product (e.g., api_key_id, project_id for inference); GPU cluster breakdown by cluster_id inferred but not explicitly documented in API examples.", "confidence": "medium", "checked": "2026-10-01" }, "persist_scope": { "value": "folder", "source": "https://docs.together.ai/docs/cluster-storage.md", "quote": "Shared volumes (PVC): Persistent. Survives pod restarts, node reboots/migrations/recreations, cluster operations, and even cluster deletion.", "note": "Only shared volumes survives a stop; the rest of the filesystem (packages installed elsewhere) resets.", "checked": "2026-10-01" } }, "locations": [ { "country": "US", "state": null, "city": null, "provider_region": null, "self_serve": true, "choosable": true }, { "country": "FR", "state": null, "city": null, "provider_region": null, "self_serve": true, "choosable": true }, { "country": "NL", "state": null, "city": null, "provider_region": null, "self_serve": true, "choosable": true }, { "country": "SE", "state": null, "city": null, "provider_region": null, "self_serve": true, "choosable": true }, { "country": "RO", "state": null, "city": null, "provider_region": null, "self_serve": true, "choosable": true } ], "locations_source": "https://www.together.ai/blog/instant-gpu-clusters", "locations_quote": "150 MW+ available in Europe: France, Netherlands, Sweden, and Romania. ... USA: 2GW+ in the portfolio with 600MW of near-term capacity in US. ... Launch close to your users and data across 25+ cities.", "locations_checked": "2026-10-01" }