Very long inputs
Choose Ministral 3 14B
Ministral 3 14B reads up to 256k tokens per request, against 131k for gpt-oss-20b.
Choose gpt-oss-20b for the hardest reasoning tasks, agentic coding, tool use and support agents and reasoning over long documents. Choose Ministral 3 14B for very long inputs, images as input and the latest release.
Each answer below comes straight from the published prices, specifications and independent benchmark results.
Choose Ministral 3 14B
Ministral 3 14B reads up to 256k tokens per request, against 131k for gpt-oss-20b.
Choose gpt-oss-20b
gpt-oss-20b scores 9 on the Artificial Analysis Intelligence Index, against 6 for Ministral 3 14B.
Choose gpt-oss-20b
gpt-oss-20b scores 10.6% on Terminal-Bench Hard, against 4.5% for Ministral 3 14B.
Choose gpt-oss-20b
gpt-oss-20b scores 60.2% on τ²-Bench (agentic tool use), against 27.2% for Ministral 3 14B.
Choose gpt-oss-20b
gpt-oss-20b scores 34.7% on Long-context reasoning (AA-LCR), against 26.3% for Ministral 3 14B.
Choose Ministral 3 14B
Ministral 3 14B accepts images as input; gpt-oss-20b does not.
Choose Ministral 3 14B
Ministral 3 14B is the newer release (December 2025, against August 2025 for gpt-oss-20b).
| gpt-oss-20b | Ministral 3 14B | |
|---|---|---|
| Provider | OpenAI | Mistral AI |
| Category | Language models | Language models |
| Status | Generally available | Generally available |
| Headline price | No data | No data |
| Input | No data | No data |
| Output | No data | No data |
| Cached input | No data | No data |
| Context window | 131k | 256k |
| Max output | 131k | No data |
| Released | August 2025 | December 2025 |
| Knowledge cutoff | June 2024 | No data |
| Licence | Open weights | Open weights |
| Capabilities | Reasoning, Tool calling, Self-hostable | Vision, Tool calling, Self-hostable |
| API model id | gpt-oss-20b | ministral-14b-2512 |
Leader in bold.
| Benchmark | gpt-oss-20b | Ministral 3 14B |
|---|---|---|
| Intelligence Index | 9 | 6 |
| Terminal-Bench Hard | 10.6% | 4.5% |
| SciCode | 38.9% | 23.8% |
| GPQA Diamond | 68.8% | 57.2% |
| Humanity's Last Exam | 11% | 4.6% |
| IFBench (instruction following) | 65.1% | 32% |
| τ²-Bench (agentic tool use) | 60.2% | 27.2% |
| Long-context reasoning (AA-LCR) | 34.7% | 26.3% |
| MMMU-Pro (visual reasoning) | No data | 49.8% |
Independent evaluations by Artificial Analysis, run the same way for every model, at the reasoning setting shown. Percentages unless marked as an index.
Choose gpt-oss-20b for the hardest reasoning tasks, agentic coding, tool use and support agents and reasoning over long documents. Choose Ministral 3 14B for very long inputs, images as input and the latest release.
A price comparison is not possible: one of them has no comparable published price.
Ministral 3 14B reads more per request: 256k tokens against 131k.
gpt-oss-20b is made by OpenAI and Ministral 3 14B by Mistral AI. A price comparison is not possible: one of them has no comparable published price. Ministral 3 14B reads more per request: 256k tokens against 131k. Ministral 3 14B is the newer model (December 2025, against August 2025). Only gpt-oss-20b offers reasoning. Only Ministral 3 14B offers vision.
gpt-oss-20b scores 10.6% on Terminal-Bench Hard, against 4.5% for Ministral 3 14B.
Ministral 3 14B was released on December 2025 and gpt-oss-20b on August 2025.