Mistral Medium 3 vs DeepSeek-R1

It is close. Our weighted score puts them within 1 points (Mistral Medium 3 53/100, DeepSeek-R1 51/100), so choose by what matters most for your work: DeepSeek-R1 for raw capability, Mistral Medium 3 on price and Mistral Medium 3 for long inputs. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.

Mistral Medium 3 is cheaper at $0.40 input / $2.00 output per million tokens (official Mistral API price). DeepSeek-R1 costs $0.70 input / $2.60 output per million tokens (median across 11 API providers). At a typical mix of three input tokens to one output token, that is $0.80 per million tokens for Mistral Medium 3 versus $1.18 for DeepSeek-R1 (1.5× as much).

DeepSeek-R1 scores higher on the Capabilities Index (ECI): DeepSeek-R1 139.0 (#104 of 148) and Mistral Medium 3 134.1 (#117 of 148). Their confidence ranges do not overlap (136.2–140.4 vs 130.5–135.6), so the gap is a real one. On individual benchmarks: GPQA Diamond — DeepSeek-R1 71.7%, Mistral Medium 3 59.5%; OTIS Mock AIME 2024–2025 — DeepSeek-R1 53.3%, Mistral Medium 3 32.2%.

There are no published SWE-bench Verified results for Mistral Medium 3 and DeepSeek-R1 yet, so there is no like-for-like coding score. On overall capability, DeepSeek-R1 leads, which tends to carry over to coding, but test on your own codebase. Both support tool calling for agent workflows.

Mistral Medium 3 has the largest context window at 131,072 tokens, against 128,000 for DeepSeek-R1. Maximum output per response: Mistral Medium 3 up to 131,072, DeepSeek-R1 up to 32,768 tokens.

Mistral Medium 3 accepts text and images; DeepSeek-R1 accepts text. Mistral Medium 3 handles the widest range of inputs.

DeepSeek-R1 publishes its weights and can be self-hosted; Mistral Medium 3 is proprietary.

Mistral Medium 3 is the newest, released May 7, 2025. DeepSeek-R1 came out Jan 20, 2025. Knowledge cutoff: Mistral Medium 3 May 2025, DeepSeek-R1 Jul 2024.

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Oct 4, 2026.