# Lightning AI

Official pricing: https://lightning.ai/pricing  
Category: dev-env · Isolation: vm

## Pricing regimes (raw)

- **The one free Default CPU Studio (4 cores / 16 GB)** (pool)
- **On-demand CPU Studio (fixed shapes)** (sizes): Large (CPU) 8 vCPU/32 GiB $0.51/h; X-Large (CPU) 16 vCPU/64 GiB $0.99/h; Data prep 32 vCPU/128 GiB $1.48/h; Data prep max 64 vCPU/256 GiB $2.69/h; Data prep ultra 96 vCPU/384 GiB $4.79/h
- **Interruptible CPU Studio** (sizes): Default (CPU) interruptible 4 vCPU/16 GiB $0.22/h; Large (CPU) interruptible 8 vCPU/32 GiB $0.33/h; X-Large (CPU) interruptible 16 vCPU/64 GiB $0.54/h; Data prep ultra interruptible 96 vCPU/384 GiB $4.51/h [spot]
- **GPU Studio, on-demand (1-GPU rates)** (resource): $0/vCPU-h, $0/GiB-h
- **GPU Studio, interruptible (midpoint of live band)** (resource): $0/vCPU-h, $0/GiB-h [spot]

## Features

Yes: snapshots, fork/clone, pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), Docker inside, root, code interpreter, SSH, HTTPS preview URLs, egress allowlist, private networking, self-hosting, BYOC, SOC 2, HIPAA, SSO, GPU, Python, Node.js, wake on request, live fork (no pause), audit logs, EU data residency, spend limits, automatic snapshots, snapshots on demand, fast boot (benchmarked), gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes, shared volumes

No: memory snapshots, browser, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, public IPv4, static egress IP, open source, arm64, Windows, macOS, GPU desktop, HTTP method/path egress rules, live resize, memory fork, MCP server, agent harness API, hosted agent API (their own agent), secret proxy, secret proxy for any API

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/lightning-ai.json → feature_evidence

## Caveats

- The Sandbox API (cpu-1 1/1 GB, cpu-2 2/8, cpu-4 4/16, cpu-8 8/32, cpu-16 16/64) has NO published price. The card prices Studio shapes. cpu-4/8/16 match the Default/Large/X-Large Studios, but whether Studio rates apply is unverified.
- No on-demand price is published for a second Default 4/16 CPU Studio (only 'Free' for one). The engine therefore puts 4-core on-demand workloads on Large 8/32 ($0.51/h), which may overestimate.
- GPU Studio prices are per GPU with host CPU/RAM bundled. The per-GPU rate varies by 1/2/4/8-GPU shape (e.g. A100 80GB $2.71 single vs $1.55 on 8-GPU; A100 40GB $2.19 single vs $3.87 on 8-GPU); only 1-GPU rates are encoded (B200 is listed for 8-GPU nodes only at $9.86/GPU).
- Interruptible GPU bands are live and move intraday; on T4, RTX PRO 6000 and A100 80GB they sit above on-demand. TPU Studios ($3.03-$4.72/TPU-h) are not encoded.
- 10 GB free storage is account-wide (Drive), not per sandbox; above it $0.10/GB-month billed daily. Drive caps: Free 50 GB, Pro 200 GB, Teams 2 TB. Snapshot storage price not separately published.
- When credits run out, Lightning warns and then gracefully shuts down all Studios: there is no overage. Credits (free and paid) expire after 12 months, so monthly plan credits roll over within a year.
- Free plan concurrency for paid CPU Studios is not published. GPU concurrency: Free 2, Pro 6, Teams 12.
- Egress, ingress and public IPv4 pricing not published. Regions are not listed as geographies; compute runs on AWS, GCP, Lightning (Voltage Park), Lambda, Nebius, NScale and Voltage.
- Auto-sleep after 10 min idle (customisable on paid tiers), so up to 10 min idle tail is billed per session. A sleeping Studio costs $0 compute.
- 2026-09-21 reprice (third-party usagepricing.com): GPU rates moved and the Free tier went from 15 monthly credits to a one-time up-to-30 grant. usagepricing lists H100 at $4.50, but the official page shows $4.68 (2026-09-28).
- Annual plans are flagged 'sales' only because they are 1-year prepaid commitments; they are self-serve.
- Studio start/wake time is not published, so no boot overhead is modelled; Studios are dev environments, not an instant sandbox API.
- Multi-GPU per-GPU rates (live 2026-09-29): 2-GPU L4 $1.19, RTX PRO 6000 $2.82, A100 40GB $2.19, A100 80GB $2.69, H100 $4.68, H200 $4.50; 4-GPU T4 $0.88, L4 $1.20, L40S $2.97, RTX PRO 6000 $2.79, A100 40GB $2.19, A100 80GB $2.68, H100 $4.68, H200 $4.50. Multi-GPU interruptible bands are sometimes far below 1-GPU: 4-GPU T4 $0.29-0.35, 4-GPU L40S $1.90-2.28, 8-GPU L40S $2.80-3.36, 8-GPU A100 40GB $2.57-3.08 per GPU (no 1-GPU spot for L40S / A100 40GB).
- Starting a Studio from a custom launch image ('Bring your own images') is Enterprise-only; lower tiers customise the persistent environment interactively (Docker is installed inside Studios). The Sandbox API separately accepts a user OCI image.
- Other storage tiers: high-performance volumes on Lightning Cloud $0.30/GiB-month; AWS EFS $0.0125/GB-day + $0.0375/GB read + $0.075/GB write; GCP Filestore $0.5625/GB-month (EFS/Filestore Enterprise-only per compare table). The 10 GB free is per account and storage is billed while Studios sleep.
- Max concurrent GPUs (Free 2 / Pro 6 / Teams 12) counts GPUs across Studios, Batch Jobs, multi-machine training and Deployments together.

## How this provider charges


Lightning AI sells persistent **Studios** (cloud dev workspaces that can also run agent code) plus a newer
**Sandbox API** (`cpu-1` … `cpu-16`). Everything is billed per second from a **credit pool (1 credit = $1)**.
There is no per-vCPU/per-GiB rate: CPU machines are fixed shapes with a bundled hourly price, and GPU machines
are priced per GPU. The Sandbox API has **no published price**. The rates below were re-verified on
2026-09-28 by rendering https://lightning.ai/pricing in headless Chrome on every tab
(`raw/lightning_ai_pricing_tabs_2026-09-28.txt`, `raw/lightning_ai_pricing_plan_toggles_2026-09-28.txt`).

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Free CPU Studio | The one active Default CPU Studio every account gets | $0. On the Free plan it must restart every 4 h; on Pro and above it has no restarts. You can create unlimited Studios, but only one runs for free at a time | Default (CPU) 4 cores / 16 GB = **Free** | https://lightning.ai/pricing (CPU tab + FAQ) |
| On-demand CPU Studio | Any extra running CPU Studio or a larger shape | Per second while running, bundled CPU + RAM | Large 8/32 GB **$0.51/h**; X-Large 16/64 **$0.99**; Data prep 32/128 **$1.48**; Data prep max 64/256 **$2.69**; Data prep ultra 96/384 **$4.79**. The on-demand price of a *second* Default 4/16 Studio is not published | https://lightning.ai/pricing (CPU tab) |
| Interruptible (spot) CPU | You opt in to interruptible machines | Per second; can be reclaimed | Default 4/16 **$0.22/h**; Large **$0.33**; X-Large **$0.54**; Data prep ultra **$4.51**; Data prep and Data prep max have no spot price | https://lightning.ai/pricing (CPU tab) |
| On-demand GPU Studio, 1 GPU | GPU workload | Per GPU-hour, per second. What CPU/RAM comes with each GPU machine is not stated | T4 $0.55, L4 $0.79, L40S $2.14, RTX PRO 6000 $2.89, A100 40GB $2.19, A100 80GB $2.71, **H100 $4.68**, H200 $4.50 | https://lightning.ai/pricing (1 GPU tab) |
| Multi-GPU (2/4/8-GPU) Studios | Multi-GPU shapes | The per-GPU rate depends on the shape and on which cloud serves it | 8-GPU: L4 $1.25, L40S $4.21, RTX PRO 6000 $2.78, A100 40GB $3.87, **A100 80GB $1.55**, H100 $4.68, H200 $4.50, B200 $9.86 per GPU | https://lightning.ai/pricing (8 GPU tab) |
| Interruptible GPU | Opt-in spot GPU | A live price band that moves during the day | T4 $0.58–0.69, L4 $0.73–0.87, RTX PRO 6000 $3.38–4.06, A100 80GB $3.66–4.39, H200 $3.82–4.58. For T4, RTX PRO 6000 and A100 80GB the **whole band is above on-demand**, even though the Free card says "80% off" | https://lightning.ai/pricing |
| TPU Studios | GCP TPUs | Per TPU-hour | 1 TPU $3.05, 4/8 TPU $3.03, 8 TPU 95 GB $4.72; spot $1.73–2.21 | https://lightning.ai/pricing (TPU tab) |
| Reserved clusters | Guaranteed capacity (A100/H100/H200/B200, 8 GPUs per machine, date range) | Quote only. Docs describe it as "reduced rates" | null | https://lightning.ai/pricing (Clusters tab), GPU marketplace docs |
| Sandbox API (`cpu-1` 1/1 GB, `cpu-2` 2/8, `cpu-4` 4/16, `cpu-8` 8/32, `cpu-16` 16/64) | Programmatic agent sandboxes | Price **not published**. The shapes cpu-4/8/16 match the Default/Large/X-Large Studios, so the Studio rates probably apply, but this is unverified | null | https://lightning.ai/docs/platform/build/sandbox.md |
| Free plan | Default plan | $0 fee. One-time credit of up to **30 credits** (5 at signup + 25 after adding a card), which expires after 12 months. When credits run low, Lightning warns you and then **gracefully shuts down all workloads**; there is no overage billing. The CPU table lists paid CPU/GPU usage as "Pay-as-you-go" | 32-core max CPU per Studio; 1 GPU per Studio; 2 concurrent GPUs; A100/H100/H200 sessions limited to 4 h; Drive limit 50 GB | https://lightning.ai/pricing, https://lightning.ai/docs/overview/faq/billing |
| Pro, monthly | Self-serve | $50/mo buys **40 credits/mo**, so $10/mo is a pure fee | 64-core CPU; 4 GPUs per Studio; 6 concurrent GPUs; 200 GB Drive; no Free-Studio restarts | pricing page, Monthly toggle |
| Pro, annual | 1-year prepay | $20/mo ($240/yr) buys **240 credits/yr**, so the fee is fully converted to credit | same limits as Pro monthly | pricing page, Annual toggle |
| Pro, academic | Academic toggle (eligibility required) | $10/mo billed annually buys 100 credits/yr (about $8.33/mo) | same limits as Pro | pricing page, Academic toggle |
| Teams, monthly | Per user | $140/user/mo buys 50 credits/user/mo | 96-core CPU on the plan card (the comparison table says 192); 8 GPUs per Studio; 12 concurrent GPUs; 2 TB Drive; spending limits; AWS Marketplace billing | pricing page |
| Teams, annual | Per user, 1-year | $119/user/mo buys 600 credits/user/yr (= 50/mo) | same limits as Teams monthly | pricing page |
| Enterprise | Sales | Custom credits at bulk pricing; use your own AWS/GCP credits; VPC/BYOC; on-prem; B200 nodes; 99.95% SLA; SAML/SSO; bring your own images | null | pricing page |
| Credits expiry | All purchased and plan credits | Credits expire 12 months after purchase, so **monthly plan credits roll over for up to a year** | — | https://lightning.ai/docs/overview/faq/billing |
| Drive storage | Data above 10 GB (account Drive) | $0.10/GB-month, billed daily; data connections (S3/GCS/R2) are not billed | Drive caps: Free 50 GB, Pro 200 GB, Teams 2 TB, Enterprise unlimited | billing FAQ |
| Sleeping Studio | 10 min idle leads to auto-sleep (customisable on paid tiers) | Compute $0 while asleep; "sleeping Studios don't cost anything" (Drive storage above 10 GB still bills) | 10 min idle tail billed | https://lightning.ai/docs/platform/build/ai-studio.md |
| Egress / IPv4 | — | Not published. Lightning Storage cloud folders are marketed as having no ingress/egress fees | null | features research |

Dated changes: on 2026-09-21 Lightning repriced its GPU rate card (third-party usagepricing.com: T4 $0.19 to $0.55, H200
$6.53 to $4.50) and **turned the Free tier's monthly 15 credits into a one-time grant of up to 30**. Plan prices did
not change. The usagepricing.com capture lists H100 at $4.50, but the live official page on 2026-09-28 shows **$4.68**
on both the 1-GPU and 8-GPU tabs. We recorded the official figure.

## Gotchas

1. **No 4 vCPU / 8 GiB on-demand shape.** The only 4-core shape (Default 4/16) is free for one Studio. The on-demand price of
   a second one is not published, so the smallest *priced* on-demand shape is Large 8/32 at $0.51/h, which is 2x the cores you asked for.
   Interruptible Default 4/16 at $0.22/h is the cheap route, but it can be interrupted.
2. **The monthly Pro fee is not all credit.** Monthly Pro costs $50 and returns 40 credits, so $10 is a pure fee. Annual Pro ($20/mo) is 100% credit.
   Teams monthly returns $50 of $140 per seat, so $90/seat is a pure fee.
3. **Spot can cost more than on-demand.** On T4, RTX PRO 6000 and A100 80GB, the interruptible band sits above the on-demand rate.
4. **The per-GPU price depends on the node shape.** An A100 80GB costs $2.71 on a 1-GPU Studio but $1.55/GPU on an 8-GPU node. An A100 40GB costs $2.19
   on a 1-GPU Studio but $3.87 on an 8-GPU node.
5. **Running out of credits means shutdown, not overage.** Lightning gracefully shuts down all Studios, which can kill a long agent session.
   Auto-reload is offered on Teams and above.
6. **The idle tail is billed.** Auto-sleep starts only after 10 min idle, and a sleeping Studio is free.
7. **The Sandbox API has no rate card**, so every number here is Studio pricing.
8. **The 10 GB free storage is account-wide** (the Drive), not per sandbox.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 sandbox-hours**, 30% CPU,
50 GiB snapshots retained, 100 GiB egress.

- The 30% CPU utilisation does not matter, because billing is wall-clock on a fixed shape.
- One of the 50 runs on the free Default Studio (176 h, $0). The other 49 need a priced shape:
  - on-demand Large 8/32: 8,624 h × $0.51 = **$4,398.24**
  - interruptible Default 4/16: 8,624 h × $0.22 = **$1,897.28**
- Storage: (50 − 10) GiB × $0.10 = **$4.00**
- Egress: not published (null)

| Regime | Monthly total |
|---|---|
| Free plan, pay-as-you-go credits, on-demand Large | **$4,402.24**. Feasible only if Free allows 49 concurrent paid CPU Studios (no cap is published). 8 h sessions are fine because the 4 h limit applies only to the free Studio and to A100/H100/H200. |
| Pro monthly | $4,402.24 + $50 − $40 credit = **$4,412.24** |
| Pro annual | **$4,402.24**, because the $20/mo fee is fully credit (1-year prepay) |
| Teams monthly (1 seat) | $4,402.24 + $140 − $50 = **$4,492.24** |
| Interruptible Default 4/16 (any plan) | $1,897.28 + $4 = **$1,901.28**. Interruptions are possible, and 8 h sessions may be cut short. |
| Sandbox API `cpu-4` | unknown (no published rate) |