Very long inputs
Choose Mistral Medium 3.5
Mistral Medium 3.5 reads up to 256k tokens per request, against 128k for Command A+.
Choose Command A+ for the latest release. Choose Mistral Medium 3.5 for very long inputs, agentic coding, tool use and support agents, reasoning over long documents and self-hosting and data control.
Each answer below comes straight from the published prices, specifications and independent benchmark results.
Choose Mistral Medium 3.5
Mistral Medium 3.5 reads up to 256k tokens per request, against 128k for Command A+.
Either one
They are practically level on the Artificial Analysis Intelligence Index (13.1 and 14.2).
Choose Mistral Medium 3.5
Mistral Medium 3.5 scores 33.3% on Terminal-Bench Hard, against 25% for Command A+.
Choose Mistral Medium 3.5
Mistral Medium 3.5 scores 94.2% on τ²-Bench (agentic tool use), against 80.7% for Command A+.
Choose Mistral Medium 3.5
Mistral Medium 3.5 scores 69.3% on Long-context reasoning (AA-LCR), against 52.7% for Command A+.
Choose Mistral Medium 3.5
Mistral Medium 3.5 has open weights, so you can run it on your own servers; Command A+ is only available through its provider.
Choose Command A+
Command A+ is the newer release (May 2026, against April 2026 for Mistral Medium 3.5).
| Command A+ | Mistral Medium 3.5 | |
|---|---|---|
| Provider | Cohere | 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 | 128k | 256k |
| Max output | No data | No data |
| Released | May 2026 | April 2026 |
| Knowledge cutoff | No data | No data |
| Licence | Proprietary | Open weights |
| Capabilities | Vision, Tool calling, Structured output | Vision, Tool calling, Structured output, Self-hostable |
| API model id | command-a-plus-05-2026 | mistral-medium-3-5 |
Leader in bold.
| Benchmark | Command A+ | Mistral Medium 3.5 |
|---|---|---|
| Intelligence Index | 13.1 | 14.2 |
| Terminal-Bench Hard | 25% | 33.3% |
| SciCode | 38.5% | 40.2% |
| GPQA Diamond | 76.1% | 74.8% |
| Humanity's Last Exam | 12% | 13.8% |
| IFBench (instruction following) | 73.9% | 68.8% |
| τ²-Bench (agentic tool use) | 80.7% | 94.2% |
| MMMU-Pro (visual reasoning) | 63.2% | 64.9% |
| Long-context reasoning (AA-LCR) | 52.7% | 69.3% |
Independent evaluations by Artificial Analysis, run the same way for every model, at the reasoning setting shown. Percentages unless marked as an index.
Choose Command A+ for the latest release. Choose Mistral Medium 3.5 for very long inputs, agentic coding, tool use and support agents, reasoning over long documents and self-hosting and data control.
A price comparison is not possible: one of them has no comparable published price.
Mistral Medium 3.5 reads more per request: 256k tokens against 128k.
Command A+ is made by Cohere and Mistral Medium 3.5 by Mistral AI. A price comparison is not possible: one of them has no comparable published price. Mistral Medium 3.5 reads more per request: 256k tokens against 128k. Command A+ is the newer model (May 2026, against April 2026). Only Mistral Medium 3.5 offers self-hostable.
Mistral Medium 3.5 scores 33.3% on Terminal-Bench Hard, against 25% for Command A+.
Command A+ was released on May 2026 and Mistral Medium 3.5 on April 2026.