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. | 7B/13B/33B/65B | — |
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. | 2k | 128k |
| These models have no shared benchmark scores. | ||
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | — | 61.6% |
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. | 23.7% | — |
MATHReasoning — accuracy on the original competition mathematics test set | 10.6% | — |
ARC-ChallengeReasoning — multiple-choice science questions from the AI2 Reasoning Challenge; unrelated to ARC-AGI | 56% | — |
HellaSwagReasoning — commonsense completion of short scenarios | 84.2% | — |
WinoGrandeReasoning — resolving ambiguous pronouns in context | 77% | — |
MMLUGeneral knowledge — A 57-subject multiple-choice exam — history, law, medicine, maths — that was the standard measure of how much a model knows from 2020 until roughly 2024, when frontier scores crowded into the high 80s and labs moved on to harder tests. The scores here were published years apart under different testing setups, so read them as a historical record rather than a like-for-like ranking. Higher is better. | 63.4% | — |
GSM8KGrade-school math — Grade-school maths word problems that take a few steps of arithmetic to work through. It separated the models of 2022 and 2023 sharply, then saturated. One caveat on the historical numbers: OpenAI included part of the GSM8K training set in GPT-4's pre-training mix, so GPT-4's score is not a clean few-shot result. Higher is better. | 50.9% | — |
| Overview | ||
| Company | Meta | Mistral |
| Release date | Feb 24 2023 | Jul 10 2025 |
| Access | Open Weight | Closed |
| Model details | View model | View model |
Other comparisons
LLaMA 1vsClaude Fable 5.2Devstral MediumvsClaude Fable 5.2LLaMA 1vsGPT-6.2 AstraDevstral MediumvsGPT-6.2 AstraLLaMA 1vsGemini 4 HydrogenDevstral MediumvsGemini 4 HydrogenLLaMA 1vsGrok 4.7Devstral MediumvsGrok 4.7LLaMA 1vsDeepSeek-V4.1-Flash-1017Devstral MediumvsDeepSeek-V4.1-Flash-1017LLaMA 1vsKimi K3.1 ProDevstral MediumvsKimi K3.1 ProFrequently asked questions
LLaMA 1 and Devstral Medium don't publish scores on any of the same benchmarks, so there's no direct head-to-head comparison. LLaMA 1 shipped 867 days before Devstral Medium, so benchmark comparisons should account for the intervening progress.
Context windows are 2k (LLaMA 1) vs 128k (Devstral Medium). LLaMA 1 is open weight, while Devstral Medium is closed.
Direct benchmark comparisons are unavailable — LLaMA 1 and Devstral Medium don't publish scores on any of the same benchmarks.