Free GPT-OSS API.
No key. No subscription.
OpenAI's GPT-OSS — the only open-weights models OpenAI ever released. The 20B variant, 128K context, hosted free on NVIDIA and called through BlockRun. (NVIDIA retired its free 120B hosting on 2026-09-03; calls to it now redirect.)
Try it now.
No API key. No wallet. No signup. Paste this into any terminal — the response streams back from GPT-OSS 20B, routed through BlockRun.
curl https://blockrun.ai/api/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "nvidia/gpt-oss-20b",
"messages": [{"role": "user", "content": "Explain prompt caching in two sentences"}]
}'- Context
- 128K
- Price
- free
- Best for
- coding
6 ways to use GPT-OSS free.
BlockRun is the access layer. Pick the surface that matches how you build — terminal, notebook, IDE, agent runtime — and the same free models work everywhere.
- shell
curl https://blockrun.ai/api/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{ "model": "nvidia/gpt-oss-20b", "messages": [{"role": "user", "content": "Explain prompt caching in two sentences"}] }' - python
# Works with the OpenAI SDK — no key required for free models from openai import OpenAI client = OpenAI( base_url="https://blockrun.ai/api/v1", api_key="not-needed-for-free-models", ) response = client.chat.completions.create( model="nvidia/gpt-oss-20b", messages=[{"role": "user", "content": "Explain prompt caching in two sentences"}], ) print(response.choices[0].message.content) - 03
ClawRouter
smart router for OpenClaw / Claude Code — auto-picks free models when possible
Learn more →shell# Install once npm install -g @blockrun/clawrouter # Then point any OpenAI-compatible client at the local proxy. # ClawRouter routes to nvidia/gpt-oss-20b (or the cheapest capable model) # without changing your code. - typescript
// Works with the OpenAI SDK — no key required for free models import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://blockrun.ai/api/v1", apiKey: "not-needed-for-free-models", }); const r = await client.chat.completions.create({ model: "nvidia/gpt-oss-20b", messages: [{ role: "user", content: "Explain prompt caching in two sentences" }], }); console.log(r.choices[0].message.content); - 05
Claude Code MCP
19 tools for Claude Code, Cursor & ChatGPT — call any free model from inside your editor
Learn more →shell# Add the BlockRun MCP server (Claude Code, Cursor, or ChatGPT desktop) claude mcp add blockrun -s user -- npx -y @blockrun/mcp@latest # Then call from inside the editor: # blockrun_chat(model="nvidia/gpt-oss-20b", messages=[{role:"user", content:"…"}]) - 06
Franklin Agent
the AI agent with a wallet — free OSS models for routine tasks, paid models on demand
Learn more →shell# Install Franklin Agent npm i -g @blockrun/franklin # Run with this model franklin start --model nvidia/gpt-oss-20b -p "Summarize the README"
We don't share
your data.
Your prompt goes to the model that answers it, and nowhere else. No training, no resale, no profile linking.
- No training, no resale. Your prompt goes only to the model that answers it.
- Wallet in, prompt out. Pseudonymous by default — no email, no phone number, no identity documents.
- Read the code, audit the wire format, run it yourself. @blockrun/llm and blockrun-llm on npm and PyPI.
Want Claude, GPT-5,
or Gemini too?
No subscription. No monthly minimum. Pay per call in USDC via x402 — works the same endpoint, same SDK, same model IDs. Connect a wallet, top up $5, call any frontier model. No credit card.
Everything you might
be wondering.
- Is this a real OpenAI model?
- Yes — GPT-OSS is OpenAI's open-weights release. The model files are public; NVIDIA hosts inference free on build.nvidia.com, and BlockRun routes calls to it without auth.
- How does it compare to GPT-5?
- GPT-OSS 20B is closer to GPT-3.5 class than current frontier — but it's free, fast, and runs on a permissive license. For frontier quality, use GPT-5.5 on the paid tier.
- Why is it not in the public catalog?
- Hidden from /v1/models for privacy reasons (NVIDIA's free tier may use prompts for service improvement). Still callable by ID for legacy / direct integrations.
- What happened to GPT-OSS 120B?
- NVIDIA retired its free hosting of gpt-oss-120b on 2026-09-03 (the upstream now answers 410 Gone). Calls to nvidia/gpt-oss-120b — and to the bare name gpt-oss-120b — are redirected to nvidia/nemotron-3-super-120b, a free 120B-class model, so pinned integrations keep working. The 20B is unaffected.