gpt-oss-120b
OpenAI · released Aug 4, 2025 · openai/gpt-oss-120b
- Type
- Open weightsApache License 2.0
- Params
- 120B
- Context
- 131K
5.1B active per word · about 98K words of context
Our take
Written Sep 30, 2026gpt-oss-120b is a downloadable text model you can run yourself, because only a fraction of its parameters work on each token. Its licence allows commercial use, changes and redistribution, and its measured quality sits mid-field to lower-field on the preference boards.
Pick it when you want a large model on your own machine without a licence getting in the way, or for long-document work where a report or a stack of files need not be split up first. The request capacity leaves room for all of it, though reliable recall across a long input is unverified in our data. Skip it if you need a model near the top of the preference boards, or one that resolves real GitHub issues without review.
The case for it
- About 5.1 billion of its 120.4 billion parameters work per token, so memory in use is closer to a small model than to a 120-billion-parameter one.
- The licence allows commercial use, changes and redistribution (Apache License 2.0).
- A long report or a stack of documents fits beside the question without being split up first, though room to hold it is not a guarantee of accurate recall.
The case against it
- Mid-field to lower-field on the preference boards: 122nd of 168 on Arena Text (overall) and 138th of 168 on Arena Creative Writing as of 25 Sep 2026, which record which answer people preferred rather than whether it was correct.
- Weak on real software-engineering work: 26% of real GitHub issues resolved end-to-end on SWE-bench Verified inside the mini-SWE-agent harness, 36th of 42 as of 19 Feb 2026 — a result for the model inside that harness, not for the model alone.
How good is it?
An open text model for everyday questions and code, though its drafting and prose trail most models.
- drafts, rewrites and editingArena Creative Writing · 138th of 168
EverydayGeneral questions and everyday reasoning
Arena Text (overall)122nd of 168 · 1352
CodingWriting and fixing code on its own
Arena Coding126th of 168 · 1390
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on SWE-bench Verified, in 36th of 42 with 26.
WritingDrafting and rewriting prose
Arena Creative Writing138th of 168 · 1277
Arena Creative Writing is the only board that has scored it for this.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done.
Every published score for this model7 scoresEvery figure we hold, from 7 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M1 Ultra (64-core GPU) · 128 GB
Room to spare. 16.5 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 2 hours and 9 days ago — each listing carries its own date.
Some hosts sell this model at two prices: on their own price list (“direct”) and on their OpenRouter listing (“through OpenRouter”). Where the two differ, the row shows both, each with the date we last read it.
Google Vertex AI, through OpenRouter
Cheapest of the 9 listings we can compare like for like — at 131K of context, out of 23 in the table below. 7 cheaper rows there are outside that comparison: a different quantisation.
- per 1M tokens
- $0.090 in / $0.36 out
- Context served
- 131K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| CoreWeavefp4Through OpenRouter | $0.030 / $0.17checked 2 hours ago | 131K118K max reply | 34 tok/s | No | No | Confirmed |
| DeepInfrabf16Direct and through OpenRouter | $0.037 / $0.17checked 2 hours ago | 131K118K max reply through OpenRouter | 54 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| DekaLLMbf16Through OpenRouter | $0.030 / $0.18checked 2 hours ago | 131K118K max reply | 17 tok/s | No | No | Confirmed |
| AkashMLbf16Through OpenRouter | $0.037 / $0.19checked 2 hours ago | 131K118K max reply | 80 tok/s | No | No | Confirmed |
| Crusoebf16Through OpenRouter | $0.050 / $0.25checked 2 hours ago | 131K118K max reply | 143 tok/s | No | No | Confirmed |
| Novita AIfp4Direct and through OpenRouter | $0.050 / $0.25checked 2 hours ago | 131K33K max reply through OpenRouter | 118 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Mancer 2fp8Through OpenRouter | $0.045 / $0.28checked 2 hours ago | 131K118K max reply | 74 tok/s | No | No | Confirmed |
| Google Vertex AIglobalThrough OpenRouter | $0.090 / $0.36checked 9 days ago | 131K118K max reply | not measured | No | No | Confirmed |
| DigitalOcean GradientThrough OpenRouter | $0.060 / $0.42checked 2 hours ago | 128K4K max reply | 38 tok/s | No | No | Confirmed |
| Basetenfp4Through OpenRouter | $0.10 / $0.50checked 2 hours ago | 128K115K max reply | 138 tok/s | No | No | Confirmed |
| SambaNovaDirect and through OpenRouter | $0.22 / $0.59directchecked 2 hours ago$0.14 / $0.95through OpenRouterchecked 2 hours ago | 131K118K max reply through OpenRouter | 393 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| OpenRouterOpenRouter's own listing | $0.15 / $0.60checked 3 days ago | 131K | not measured | Unknown | Unknown | Unknown |
| Together AIThrough OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K118K max reply | 89 tok/s | No | No | Confirmed |
| Amazon BedrockThrough OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K118K max reply | 122 tok/s | No | No | Confirmed |
| Amazon Bedrockeu-west-1Through OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K118K max reply | 138 tok/s | No | No | Confirmed |
| GroqThrough OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K66K max reply | 248 tok/s | No | No | Confirmed |
| DeepInfraturbo tierbf16Through OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K16K max reply | 138 tok/s | No | No | Unknown |
| SiliconFlowfp8Through OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K8K max reply | 40 tok/s | No | No | Confirmed |
| Nebius AI Studiofp4Through OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K118K max reply | 197 tok/s | No | No | Confirmed |
| PhalaThrough OpenRouter | $0.15 / $0.60checked 2 hours ago | 131K118K max reply | 112 tok/s | No | No | Confirmed |
| Cerebrasfp16Through OpenRouter | $0.35 / $0.75checked 2 hours ago | 131K41K max reply | 652 tok/s | No | No | Confirmed |
| MaraThrough OpenRouter | $0.15 / $0.75checked 44 hours ago | 131K118K max reply | 228 tok/s | No | No | Confirmed |
| Parasailfp4Through OpenRouter | $0.10 / $0.75checked 2 hours ago | 131K118K max reply | 106 tok/s | No | No | Confirmed |
DeepInfra
Novita AI
DigitalOcean Gradient
SambaNova
OpenRouter
Together AI
Amazon Bedrock
Groq
Nebius AI Studio
CerebrasAcross the 23 listings we hold: 22 say they do not train on prompts (3 of them only through OpenRouter), 0 say they do and 1 does not say. 21 appear in the zero-retention registry we check (3 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| CoreWeavefp4Through OpenRouter | ✓ | ✓ | ✓ |
| DeepInfrabf16Direct and through OpenRouter | ✓ | ✓ | ✓ |
| DekaLLMbf16Through OpenRouter | ✓ | ✓ | ✓ |
| AkashMLbf16Through OpenRouter | ✓ | ✓ | ✓ |
| Crusoebf16Through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIfp4Direct and through OpenRouter | ✓ | ✓ | ✗ |
| Mancer 2fp8Through OpenRouter | ✓ | ✓ | ✓ |
| Google Vertex AIglobalThrough OpenRouter | ✗ | ✓ | ✓ |
| DigitalOcean GradientThrough OpenRouter | ✗ | ✗ | ✗ |
| Basetenfp4Through OpenRouter | ✓ | ✓ | ✓ |
| SambaNovaDirect and through OpenRouter | ✓ | ✗ | ✗ |
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| Together AIThrough OpenRouter | ✓ | ✓ | ✓ |
| Amazon BedrockThrough OpenRouter | ✗ | ✗ | ✗ |
| Amazon Bedrockeu-west-1Through OpenRouter | ✗ | ✗ | ✗ |
| GroqThrough OpenRouter | ✓ | ✓ | ✓ |
| DeepInfraturbo · bf16Through OpenRouter | ✓ | ✓ | ✓ |
| SiliconFlowfp8Through OpenRouter | ✗ | ✓ | ✓ |
| Nebius AI Studiofp4Through OpenRouter | ✓ | ✓ | ✓ |
| PhalaThrough OpenRouter | ✓ | ✓ | ✓ |
| Cerebrasfp16Through OpenRouter | ✓ | ✓ | ✓ |
| MaraThrough OpenRouter | ✓ | ✓ | ✓ |
| Parasailfp4Through OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 18 of 23 listings say yes, 5 say no. JSON output: 19 of 23 listings say yes, 4 say no. Strict schema: 18 of 23 listings say yes, 5 say no.
Models people weigh against gpt-oss-120b
When we formed this view
Recent changes
What moved
input −10% ($0.050 → $0.045 per 1M tokens), output −8% ($0.300 → $0.275 per 1M tokens)What moved
input +12% ($0.033 → $0.037 per 1M tokens), cache read +12% ($0.033 → $0.037 per 1M tokens)What moved
input +10% ($0.030 → $0.033 per 1M tokens), output +10% ($0.170 → $0.187 per 1M tokens), cache read +10% ($0.030 → $0.033 per 1M tokens)What moved
output −25% ($0.40 → $0.30 per 1M tokens)Each date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 23 listings does not say whether it trains on prompts, and 3 answer only through OpenRouter, not for their own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- openai/gpt-oss-120b
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- openai-gpt-oss-120b