Throughput (tokens / sec)
avg. 218.4tokens/s
o4-mini-class reasoning for tool-driven coding
Run instantly. Pay only for what you use.
GPT OSS 120B is OpenAI's most capable open-weight model, released under Apache 2.0: a 117B-parameter MoE with 5.1B active per token and a 131K-token context window, trained on the Harmony response format with configurable reasoning effort — low, medium, or high.
The effort dial is the point: AIME 2024 climbs from 56.3% at low to 95.8% at high effort, so one deployment serves quick routine calls and hard reasoning alike — while 5.1B-active routing keeps it among the fastest open models served.
At high effort it posts 80.1% on GPQA Diamond and 62.4% on SWE-bench Verified, with Harmony-trained function calling and structured JSON output — OpenAI's published evals place it near o4-mini on core reasoning.
5.1Bactive of 117B total
MoE routing makes it one of the fastest open models served at scale.
3reasoning effort levels
Low, medium, and high — AIME 2024 jumps from 56.3% to 95.8% as effort rises.
131Ktoken context window
Fits long documents and full tool-call transcripts in a single request.
GPT OSS 120B is strongest where request difficulty varies — reasoning APIs, analysis endpoints, and tool-calling backends that need depth on some calls and speed on the rest.
One dial per request — AIME 2024 climbs 56.3 → 80.4 → 95.8 across low, medium, and high effort.
80.1% on GPQA Diamond at high effort — analysis that holds up in technical and scientific domains.
Top-3 output speed among open models on Artificial Analysis, with Harmony-trained function calling.
Strict JSON schemas and tool arguments that parse the first time — built for API backends.
Reasoning APIs with mixed difficulty
Per-request effort control — low for the routine 95%, high for the hard 5%. One model, one deployment.
Math, science, and technical analysis
95.8% on AIME 2024 and 80.1% on GPQA Diamond at high effort — competition-grade reasoning on demand.
Latency-sensitive tool calling
Top-3 output speed among open models measured by Artificial Analysis, with Harmony-trained function calling.
Structured outputs at volume
Strict JSON schemas and tool calls that parse the first time, at a 5.1B-active price point.
Repo-scale coding agents
62.4% on SWE-bench Verified trails dedicated coding models — use DeepSeek V4 Flash or Qwen 3.6 27B for issue-to-patch work.
Image, audio, or video inputs
GPT OSS 120B is text-only. Route visual inputs to a multimodal model such as Qwen 3.6 35B-A3B or Gemma 4 31B.
Tight output-token budgets at high effort
High effort produces long reasoning traces — about 1.6× the median output tokens in Artificial Analysis testing. Use low or medium effort for cost-sensitive replies.
Compare Entrim’s pricing, powered by an optimized inference runtime, against other providers using the same token counts per request.
avg. 218.4tokens/s
avg. 420ms
95.8%AIME 2024
Competition mathematics measuring multi-step reasoning.
80.1%GPQA Diamond
Graduate-level scientific reasoning and analysis.
62.4%SWE-bench Verified
Real GitHub issues resolved end-to-end in agentic runs.
Code examples
curl -X POST "https://api.entrim.ai/v1/chat/completions" \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $ENTRIM_API_KEY" \ -d '{ "model": "openai/gpt-oss-120b", "messages": [ { "role": "user", "content": "Why do they call it a building if it is already built?" } ] }'SDKs & docs
Use familiar SDK patterns with quickstart examples for common production setups.
OpenAI-compatible
Keep the request format your team already knows.
No cold starts
Production requests are served without model spin-up delays.
Low-latency reasoning on a tiny budget
High-throughput agentic work with 1M context
Dense agentic coding with thinking control