gpt-oss-120b

OpenAI released 2025-08-05 apache-2.0

$0.070 per million tokens, balanced

Solar Pro 4 is both better and cheaper.

It scores +17.5 higher and costs 25% less ($0.052/M against $0.070/M) on this workload — and it does everything this model does.

Ling-3.0-flash is cheaper still (55% less) but drops 131K → 33K max output.

Our take

editorial — not a measurement

OpenAI's Apache-2.0 open-weight model, and at roughly $0.037/$0.17 on OpenRouter's default endpoint it is one of the cheapest credible models available. The strategic value is licence, not benchmark: Apache-2.0 with no field-of-use restriction means you can self-host, fine-tune and ship it commercially without asking anyone. Treat listed prices as one host's rate — Groq, Fireworks, Together and DeepInfra all price the same weights differently, and self-hosting changes the economics entirely.

Strengths

  • Apache-2.0 — genuinely unrestricted commercial use
  • ~$0.037/$0.17 at the default hosted endpoint
  • 131k context window
  • Self-hostable, so no vendor lock-in or rate limits

Weaknesses

  • Well below frontier capability
  • Text-only
  • Quoted price is one host's rate, not a canonical price

Reach for it when

  • Cost-floor workloads
  • On-prem or air-gapped deployment
  • Fine-tuning bases where licence matters

Avoid it if

  • You need frontier reasoning
  • You need multimodal input

Sources: openrouter.ai

Every price dimension

Input$0.037/M
Output$0.170/M

Independent scores

Intelligence24.1
Coding30.4
Agentic13.4
LMArena Elo1365.6default

Sources are listed separately rather than averaged. Across the 62 models both have scored they correlate at r = 0.806 — close agreement overall, but four models rank very differently between them, and a blended score would hide exactly those.

Capability

Context window131K
Max output131K
Input modestext
Tool useyes
Reasoningalways on
Knowledge cutoff2024-06-30
Open weightsyes

Provenance

Price sourceopenrouter.ai
Fetched2026-08-24
Quality dataverified
Cross-checkedvendor page

gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized...

Markdown for LLMs