The Open-Source AI Stack
RSS
All models

Models · Compare

Qwen 3 235B A22B Instruct vs Mistral Medium 3

Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.

Specs

Field A: Qwen 3 235B A22B Instruct B: Mistral Medium 3
Released 2025-04-282025-05-07
Developer AlibabaMistral AI
Openness OpenProprietary
License Apache-2.0Proprietary
OSI-approved yesno
Data released nono
Training code nono
Architecture moeunknown
Total params 235B
Active params 22B
Experts 128 (8 active)
Context window 131K128K
Attention gqaunknown
Position enc. ropeunknown
Pretraining tokens 36.0T
Post-training sft, dpo, grposft, rlhf
Training hardware
$/M input $0.45$0.40
$/M output $1.80$2.00
Output tok/sec 66.629

Benchmarks

Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.

General reasoning

MMLU-Pro 83.0 2025-04-28 76.0 2026-05-21
GPQA-Diamond 57.8 2026-05-21

Code

LiveCodeBench 34.3 2026-05-21 40.0 2026-05-21

Math

MATH 90.2 2026-05-21 90.7 2026-05-21
AIME 2024 85.7 2025-04-28 44.0 2026-05-21
AIME 2025 23.7 2026-05-21 30.3 2026-05-21

Context · A

Qwen's first major MoE release, with a hybrid thinking-vs- non-thinking inference mode controllable per request. Apache 2.0 across the size ladder reset the openness baseline among Chinese labs.

Context · B

Mid-tier flagship released May 7 2025 at $0.40 / $2.00 per Mtok with a 128K context window. Mistral positioned it as roughly 90 percent of Claude Sonnet 3.7 performance at a fraction of the cost, with deployment supported on self-hosted setups starting at four GPUs.

Qwen 3 235B A22B Instruct detail → · Mistral Medium 3 detail →