gpt-oss-120b
OpenAI · released Aug 4, 2025 · openai/gpt-oss-120b
- Type
- Open weightsApache License 2.0
- Params
- 120B
- Context
- 131K
about 98K words of context
Our take
Written Aug 3, 2026gpt-oss-120b is OpenAI's first downloadable model, released in 2025 with a permissive Apache licence and a 120.4-billion-parameter dense architecture. It offers broad measured coverage across chat, coding and creative tasks, though its scores vary sharply by category and it is text-only.
Pick this as a research or fine-tuning foundation where a permissive licence matters — Apache 2.0 allows commercial use, redistribution and modification. Use it for coding workloads where its measured score is strongest, or where host choice lets you trade cost against speed. Skip it if you need multimodal handling, if creative writing quality is critical, or if you want sparse-parameter efficiency.
The case for it
- Broad benchmark coverage with consistent scores across two evaluation dates — most categories drifted less than 0.2 points between measurements.
- Truly permissive open licence from a major frontier lab: commercial use, fine-tuning, redistribution and modification all permitted.
- Ten distinct hosting options, with a wide spread of throughput and pricing.
- Coding performance is 37.8 points above its overall text score on the Arena leaderboard.
The case against it
- No parameter efficiency: 120.4 billion active parameters with no disclosed sparse or mixture-of-experts architecture, so inference cost scales with the full model size.
- Creative writing is a clear relative weakness, scoring 74.8 points below its coding score.
- Throughput varies 7.4-fold by provider for the same model, suggesting strong infrastructure sensitivity.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)100th of 143 · 1352.3
CodingWriting and fixing code on its own
Arena Coding103rd of 143 · 1390.1
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
gpt-oss-120b is not on Arena Agent (IPS), which is where the rating would come from, so there is no rating here. It is on SWE-bench Verified, in 33rd of 39 with 26.
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where gpt-oss-120b placed and give it no mark out of five.
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, which is why they get no rating.
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?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
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.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 23 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $0.037 in / $0.17 out
- Context served
- 131K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| DeepInfrabf16 | $0.037 / $0.17 | 131K | 40 tok/s | No | No | Confirmed |
| OpenRouter | $0.037 / $0.17 | 131K | not measured | Unknown | Unknown | Unknown |
| CoreWeavefp4 | $0.030 / $0.17 | 131K | 22 tok/s | No | No | Confirmed |
| DeepInfrabfloat16 | $0.037 / $0.17 | 131K | not measured | Unknown | Unknown | Unknown |
| Novita AI | $0.050 / $0.25 | 131K | not measured | Unknown | Unknown | Unknown |
| Novita AIfp4 | $0.050 / $0.25 | 131K | 102 tok/s | No | No | Confirmed |
| Google Vertex AIglobal | $0.090 / $0.36 | 131K | 210 tok/s | No | No | Confirmed |
| SiliconFlowfp8 | $0.050 / $0.45 | 131K | 18 tok/s | No | No | Confirmed |
| DigitalOcean Gradient | $0.070 / $0.49 | 128K | 34 tok/s | No | No | Confirmed |
| Mancer 2fp8 | $0.10 / $0.50 | 131K | 31 tok/s | No | No | Confirmed |
| Basetenfp4 | $0.10 / $0.50 | 128K | 157 tok/s | No | No | Confirmed |
| SambaNova | $0.22 / $0.59 | 131K | not measured | Unknown | Unknown | Unknown |
| Phala | $0.15 / $0.60 | 131K | 80 tok/s | No | No | Confirmed |
| DeepInfraturbo tierbf16 | $0.15 / $0.60 | 131K | 129 tok/s | No | No | Unknown |
| Together AI | $0.15 / $0.60 | 131K | 148 tok/s | No | No | Confirmed |
| Amazon Bedrock | $0.15 / $0.60 | 131K | 190 tok/s | No | No | Confirmed |
| Amazon Bedrockeu-west-1 | $0.15 / $0.60 | 131K | 122 tok/s | No | No | Confirmed |
| Groq | $0.15 / $0.60 | 131K | 395 tok/s | No | No | Confirmed |
| Nebius AI Studiofp4 | $0.15 / $0.60 | 131K | 285 tok/s | No | No | Confirmed |
| Cerebrasfp16 | $0.35 / $0.75 | 131K | 919 tok/s | No | No | Confirmed |
| Parasailfp4 | $0.10 / $0.75 | 131K | 142 tok/s | No | No | Confirmed |
| Mara | $0.15 / $0.75 | 131K | 137 tok/s | No | No | Confirmed |
| SambaNova | $0.14 / $0.95 | 131K | 325 tok/s | No | No | Confirmed |
DeepInfra
Novita AI
Google Vertex AI
DigitalOcean Gradient
SambaNova
Together AI
Amazon Bedrock
Groq
Nebius AI Studio
CerebrasAcross the 23 listings we hold: 19 say they do not train on prompts, 0 say they do and 4 do not say. 18 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| DeepInfrabf16 | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
| CoreWeavefp4 | ✓ | ✓ | ✓ |
| DeepInfrabfloat16 | |||
| Novita AI | |||
| Novita AIfp4 | ✓ | ✓ | ✓ |
| Google Vertex AIglobal | ✗ | ✓ | ✓ |
| SiliconFlowfp8 | ✗ | ✓ | ✓ |
| DigitalOcean Gradient | ✗ | ✗ | ✗ |
| Mancer 2fp8 | ✓ | ✓ | ✓ |
| Basetenfp4 | ✓ | ✓ | ✓ |
| SambaNova | |||
| Phala | ✓ | ✓ | ✓ |
| DeepInfraturbo · bf16 | ✓ | ✓ | ✓ |
| Together AI | ✓ | ✓ | ✓ |
| Amazon Bedrock | ✗ | ✗ | ✗ |
| Amazon Bedrockeu-west-1 | ✗ | ✗ | ✗ |
| Groq | ✓ | ✓ | ✓ |
| Nebius AI Studiofp4 | ✓ | ✓ | ✓ |
| Cerebrasfp16 | ✓ | ✓ | ✓ |
| Parasailfp4 | ✓ | ✓ | ✓ |
| Mara | ✓ | ✓ | ✓ |
| SambaNova | ✓ | ✗ | ✗ |
Tool calling: 15 of 23 listings say yes, 5 say no, 3 publish no parameter list. JSON output: 16 of 23 listings say yes, 4 say no, 3 publish no parameter list. Strict schema: 16 of 23 listings say yes, 4 say no, 3 publish no parameter list.
Models people weigh against gpt-oss-120b
When we formed this view
Dates behind this page
Prices last checked 35h ago
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.
- 3 of 23 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 4 of 23 listings do not say whether they train on prompts.
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
- Modality record
- text->text
- Catalogue slug
- openai-gpt-oss-120b