Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 671B | — |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 128k | 128k |
| API pricingUSD per 1M tokens · lower wins | ||
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | — | $0.40Mistral |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | — | $2.00Mistral |
| — | $180Mistral · zdr | |
| Benchmarks | ||
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | 88.7% | 30.3% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 85.7% | 57.1% |
HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better. | — | 92.1% |
IFEvalAgentic tasks — following verifiable instructions in generated responses | — | 89.4% |
Humanity's Last Exam · text only, no toolsMultidisciplinary reasoning — accuracy on the 2,158 text-only questions without tools | 30.6% | — |
MATH-500Reasoning — accuracy on the 500-problem MATH subset | — | 91% |
AIME 2025Reasoning — single-answer accuracy on the 2025 invitational mathematics exam without tools | 96% | — |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | — | 77.2% |
MMMUMultimodal — Tests the AI on understanding images and text together across many college subjects. Higher is better. | — | 66.1% |
| Overview | ||
| Company | DeepSeek | Mistral |
| Release date | Dec 1 2025 | May 7 2025 |
| Access | Open Weight | Closed |
| Model details | View model | View model |
Other comparisons
DeepSeek-V3.2-SpecialevsClaude Fable 5.2Mistral Medium 3vsClaude Fable 5.2DeepSeek-V3.2-SpecialevsGPT-6.2 AstraMistral Medium 3vsGPT-6.2 AstraDeepSeek-V3.2-SpecialevsGemini 4 HydrogenMistral Medium 3vsGemini 4 HydrogenDeepSeek-V3.2-SpecialevsMuse EmberMistral Medium 3vsMuse EmberDeepSeek-V3.2-SpecialevsGrok 4.7Mistral Medium 3vsGrok 4.7DeepSeek-V3.2-SpecialevsKimi K3.1 ProMistral Medium 3vsKimi K3.1 ProFrequently asked questions
DeepSeek-V3.2-Speciale leads Mistral Medium 3 on 2 of the 2 benchmarks they both report (LiveCodeBench, GPQA Diamond). Mistral Medium 3 shipped 208 days before DeepSeek-V3.2-Speciale, so benchmark comparisons should account for the intervening progress.
Context windows are 128k (DeepSeek-V3.2-Speciale) vs 128k (Mistral Medium 3). DeepSeek-V3.2-Speciale is open weight, while Mistral Medium 3 is closed.
On LiveCodeBench, DeepSeek-V3.2-Speciale leads at 88.7% vs Mistral Medium 3 at 30.3%. On GPQA Diamond, DeepSeek-V3.2-Speciale leads at 85.7% vs Mistral Medium 3 at 57.1%.