# Thunder Compute (slim)

> Thunder Compute is a US GPU cloud renting RTX A6000, L40, A100 and H100 instances by the minute and GPU sandboxes by the second. Agents use its REST API, a hosted MCP server with OAuth, or the tnr CLI.

- Full: https://www.anchorterminal.com/tools/thunder-compute.md (~8,550 tokens) · this version ~1,880 tokens · JSON https://www.anchorterminal.com/tools/thunder-compute.json · canonical https://www.anchorterminal.com/tools/thunder-compute
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-09

**C · 56.1/100 · rank #576 of 842 · #10 in GPU & serverless compute · not agent-ready · confidence medium**


Assessment: The hosted MCP server signs in with OAuth and separate read and write scopes, and the REST API has a public OpenAPI 3.1 document with keyless price and availability endpoints. No rate limits, SLA or API changelog were found, instance creation has no idempotency key, and instances cannot be stopped, only deleted.

## Facts

- Kind: HTTP API · vendor: Thunder Compute · category: GPU & serverless compute · legal entity: Thunder GPU, Inc. · provenance 77/100
- Endpoint: `https://api.thundercompute.com:8443/v1` (HTTP, Streamable HTTP)
- Auth: OAuth or key · pricing: Pay per use · x402: no · licence: Proprietary service under Thunder Compute's Terms and Conditions. The `tnr` CLI on GitHub is MIT
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The REST API at https://api.thundercompute.com:8443/v1 (OpenAPI 3.1, 41 operations) together with the hosted MCP server, whose tool list comes from the docs because the definitions need sign-in
- MCP server: Hosted, Streamable HTTP, OAuth. 28 tools. `list_instances`, `create_instance`, `delete_instance`, `modify_instance`, `run_command`, `get_specs`, `get_availability`, `get_pricing`, `list_templates`, `list_snapshots`, `create_snapshot`, `delete_snapshot`, `list_ssh_keys`, `create_ssh_key`, `delete_ssh_key`, `add_ssh_key_to_instance`, `list_ports`, `forward_port`, `delete_port`, `get_ssh_command`, `get_scp_command`, `get_meter_data`, `get_upcoming_invoice`, `get_invoice_history`, `get_subscription`, `list_tokens`, `create_token`, `delete_token`
- Free tier: None in general. $20 of credit for students with a US university email, usable on A6000 and L40
- GPUs: RTX A6000 48 GB (one per instance), L40 48 GB, A100 80 GB and H100 80 GB in 1, 2, 4 or 8 GPU shapes, per the pricing page
- Scale to zero: Not for instances, which bill until deleted and cannot be stopped. Sandboxes end at a timeout and bill by the second
- Cold start: No figure published. The pricing page says instances are ready to SSH into in minutes
- Billing basis: Instances per minute while running, sandboxes per second, snapshots and extra disk per GB metered hourly. No egress fee
- Rate limits: Not found in the reviewed documentation
- Errors: `code`, `error` and `message`, for example 400 with `invalid_request`
- Regions: North America per the technical specifications. No region list found
- SDKs: `tnr` CLI in Go, v2.1.0 (16 September 2026), MIT. `thunder-sandbox` for Python 3.10 or newer, covering sandboxes only. VS Code and Cursor extension
- Sandboxes: Firecracker microVMs with 3 to 40 vCPUs, an optional H100 or A100, images from a registry or Dockerfile and an outbound network policy. Access is enabled for each organisation on request
- Preinstalled: CUDA 13.0, driver 580, PyTorch 2.9.0 and JupyterLab, with templates for Ollama, ComfyUI, Forge Neo and Unsloth
- Status: status.thundercompute.com on incident.io, with Website, API and Instances components
- Prices: RTX A6000 48 GB instance $0.35 per GPU-hour; L40 48 GB instance $0.79 per GPU-hour; A100 80 GB instance $1.09 per GPU-hour; H100 80 GB instance $3.20 per GPU-hour; H100 80 GB sandbox $2.95 per GPU-hour; A100 80 GB sandbox $1.95 per GPU-hour; Extra instance vCPU $0.04 per vCPU-hour; Sandbox vCPU $0.051 per vCPU-hour; Instance disk above 100 GB per GPU $0.219 per GB per month; Snapshot storage $0.05 per GB per month

- Scores: Reliability 65, Performance pending, Schema & documentation 70, Agent ergonomics 48, Security & auth 51, Payments & pricing 20, Task success pending, Maintenance & community 79, Transparency & trust 64 · total over the 7 assessed categories
- Why: Reliability, Graded on the hosted lines for the REST API and the hosted MCP server. · Schema & documentation, Public OpenAPI 3.1 document at api.thundercompute.com:8443/openapi.json with 41 operations and 72 schemas (25). · Agent ergonomics, No field selection, and `GET /instances/list` returns every instance in one map. · Security & auth, The MCP server uses OAuth 2.0 with discovery, refresh tokens and ten scopes that separate read from write for instances, snapshots and sandb… · Payments & pricing, No x402, MPP or L402 in the docs, the OpenAPI document or the pricing page (0). · Maintenance & community, The newest CLI release is v2.1.0 of 16 September 2026, and the terms and privacy policy were updated on 28 September 2026 (30). · Transparency & trust, Closed service under Terms and Conditions dated 28 September 2026 and Delaware law.
- Sources: 22, open questions: 10, both in the full twin
- Capabilities: compute.gpu, compute.containers
- JSON: https://www.anchorterminal.com/api/v1/tools/thunder-compute.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/thunder-compute.svg` or a link to https://www.anchorterminal.com/tools/thunder-compute from a page on thundercompute.com or one of its subdomains, or the README of github.com/Thunder-Compute/thunder-cli, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Call `GET /v2/status` or the `get_availability` tool before creating an instance. Availability can change before launch, and creation fails when a type is sold out.
2. List instances before retrying a failed create, because the call has no idempotency key.
3. Pass `public_key` on create. If omitted, the response carries a generated private key that is returned once.
4. To pause work, create a snapshot, delete the instance and later create a new instance from the snapshot. Snapshot storage keeps billing until deleted.
5. For headless use set `TNR_API_TOKEN` to a token from the console. The MCP server needs a browser sign-in on first connection.

## Connect

```bash
curl -fsSL https://raw.githubusercontent.com/Thunder-Compute/thunder-cli/main/scripts/install.sh | bash
```

```bash
curl https://api.thundercompute.com:8443/v2/pricing
```

```bash
claude mcp add --transport http thunder-compute https://www.thundercompute.com/mcp
```
Full config and headless snippets are in the full page. Through letme (picks today, calling later): https://letme.dev/thunder-compute

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Nebius AI Cloud | B | 67.2 | compute.gpu, compute.containers | https://www.anchorterminal.com/tools/nebius-ai-cloud.min.md |
| Baseten | B | 66.5 | compute.gpu, compute.containers | https://www.anchorterminal.com/tools/baseten.min.md |
| Hugging Face Inference Endpoints | B | 64.5 | compute.gpu, compute.containers | https://www.anchorterminal.com/tools/hugging-face-inference-endpoints.min.md |
| Modal | B | 63.6 | compute.gpu, compute.containers | https://www.anchorterminal.com/tools/modal.min.md |
| Replicate Deployments | B | 63.6 | compute.gpu, compute.containers | https://www.anchorterminal.com/tools/replicate-deploy.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)