Mistral Large 4 Preview (high): benchmark scores, price and ranking

Mistral Large 4 Preview (high) from Mistral AI ranks #374 of 827 LLMs on the LLMs Tiger aggregate benchmark score (48.8), based on 13 benchmark results captured up to 2026-11-18.

Organisation
Mistral AI
Type
Open-weight / local model
Open weights
Yes
Cheapest API price
$0.700 in / $2.10 out per million tokens (via azure_ai, observed 2026-11-18T15:47Z)
Aggregate rank
#374 of 827 (score 48.8)

Relative to other models, Mistral Large 4 Preview (high) is strongest on OTIS Mock AIME (#29 of 206) and weakest on Furniture Assembly (#35 of 36).

Across the 12 benchmarks where at least ten models are scored, Mistral Large 4 Preview (high) is in the top 10% on 0 and in the top half on 2.

Its cheapest listed input price, $0.700 per million tokens, is lower than that of 41% of the 443 models with a listed price.

Mistral Large 4 Preview (high) benchmark results

Mistral Large 4 Preview (high) benchmark scores, ranked against all models measured on each benchmark
BenchmarkScoreRankSourceCaptured
OTIS Mock AIME98.3%#29 of 206Epoch AI Benchmarking Hub2026-11-18
GPQA Diamond88.1%#72 of 302Epoch AI Benchmarking Hub2026-11-18
Chess Puzzles33.0%#43 of 84Epoch AI Benchmarking Hub2026-11-18
FrontierMath T1-358.6%#40 of 78Epoch AI Benchmarking Hub2026-11-18
GDP.pdf32.8%#19 of 37Epoch AI Benchmarking Hub2026-11-18
APEX-Agents51.6%#12 of 23Epoch AI Benchmarking Hub2026-11-18
FrontierCode39.4%#10 of 18Epoch AI Benchmarking Hub2026-11-18
FrontierMath T426.8%#40 of 60Epoch AI Benchmarking Hub2026-11-18
Mystery Games17.0%#49 of 69Epoch AI Benchmarking Hub2026-11-18
MirrorCode40.0%#25 of 35Epoch AI Benchmarking Hub2026-11-18
SimpleQA Verified20.0%#70 of 83Epoch AI Benchmarking Hub2026-11-18
WeirdML v34.2%#6 of 7Epoch AI Benchmarking Hub2026-11-18
Furniture Assembly20.8%#35 of 36Epoch AI Benchmarking Hub2026-11-18

About Mistral Large 4 Preview (high)

Mistral Large 4 Preview (high) is an open-weight model that can run on your own hardware from Mistral AI. The cheapest API listing we track charges $0.700 per million input tokens and $2.10 per million output tokens, through azure_ai. We hold 13 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where Mistral Large 4 Preview (high) stands

On our aggregate score, Mistral Large 4 Preview (high) is #374 of 827 ranked models, around the middle of the table. Its score is 48.8; the scale is centred on 50, so it sits 1.2 points below the typical model. That beats 55% of the ranked field. Among the 45 ranked Mistral AI models it is #3. The next one up is Mistral Large 4 (high) at 49.2. The next one down is codestral-2405 at 48.4.

Mistral Large 4 Preview (high), benchmark by benchmark

OTIS Mock AIME puts Mistral Large 4 Preview (high) at 98.3%, #29 of 206. That beats 86% of the models tested, but 1.7 points short of Claude Fable 5.1 (max) (100.0%). The results either side are GPT-5.6 Luna (max) (98.3%) and Grok 4.7 (xhigh) (98.1%). It covers competition maths (AIME 2024-25 style). The board median is 66.4%, so Mistral Large 4 Preview (high) is 31.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

On GPQA Diamond, Mistral Large 4 Preview (high) reaches 88.1%. It ranks #72 of 302, ahead of 76% of the field, and 8.8 points behind GPT-6.2 Astra (max) at 96.8%. The results either side are Doubao Seed 2.0 Lite (Apr 2026, high) (88.1%) and NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 (87.9%). It covers expert-level science questions. The board median is 75.1%, so Mistral Large 4 Preview (high) is 13.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on Chess Puzzles: 33.0%, #43 of 84, with 50% of the field scoring higher. Claude Fable 5.2 (max) leads at 79.0%; the closest result above is Grok 4.6 (xhigh) at 33.0%. It covers chess tactics puzzles. The median score on this board is also 33.0%. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on FrontierMath T1-3: 58.6%, #40 of 78, with 50% of the field scoring higher. Claude Fable 5.2 (max) leads at 95.1%; the closest result above is Qwen3.7 Max at 64.6%. It covers research-level maths, tiers 1-3. The board median is 61.6%, so Mistral Large 4 Preview (high) is 3.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on GDP.pdf: 32.8%, #19 of 37, with 49% of the field scoring higher. Gemini 4 Carbon (high) leads at 56.4%; the closest result above is Tencent Hy4 preview (unknown) at 35.0%. It covers economically valuable document tasks. The median score on this board is also 32.8%. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on APEX-Agents: 51.6%, #12 of 23, with 48% of the field scoring higher. Claude Sonnet 5.5 (max) leads at 75.5%; the closest result above is GPT-6 Sol (max) at 54.3%. It covers long professional agent tasks. The median score on this board is also 51.6%. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on Mystery Games: 17.0%, #49 of 69, with 70% of the field scoring higher. GPT-6.2 Astra (max) leads at 95.0%; the closest result above is Qwen3.5 397B-A17B at 18.0%. It covers deductive reasoning puzzles. The board median is 26.0%, so Mistral Large 4 Preview (high) is 9.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Mistral Large 4 Preview (high) lands below the middle on MirrorCode: 40.0%, #25 of 35, with 69% of the field scoring higher. Claude Sonnet 5.5 (max) leads at 97.8%; the closest result above is Grok 4.6 (xhigh) at 41.1%. It covers reimplementing programs from behaviour. The board median is 48.9%, so Mistral Large 4 Preview (high) is 8.9 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

SimpleQA Verified is a weak spot. Mistral Large 4 Preview (high) scores 20.0%, #70 of 83, behind 83% of the models measured. The leader, Claude Fable 5.2 (max), reaches 94.0%. It covers short factual questions. The board median is 41.4%, so Mistral Large 4 Preview (high) is 21.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

Furniture Assembly is a weak spot. Mistral Large 4 Preview (high) scores 20.8%, #35 of 36, behind 94% of the models measured. The leader, Claude Fable 5.2 (max), reaches 90.0%. It covers planning a physical assembly. The board median is 44.2%, so Mistral Large 4 Preview (high) is 23.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-11-18.

The full table above also lists the other 3 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Reasoning, maths and science: solid, ahead of 61% of the field on average across 4 benchmarks. Its best board here is OTIS Mock AIME (#29 of 206) and its weakest is FrontierMath T4 (#40 of 60). Agents and tool use: uneven, ahead of 48% of the field on average across 2 benchmarks. Its best board here is GDP.pdf (#19 of 37) and its weakest is APEX-Agents (#12 of 23). General knowledge and chat: weak, ahead of 25% of the field on average across 5 benchmarks. Its best board here is Chess Puzzles (#43 of 84) and its weakest is Furniture Assembly (#35 of 36).

Price and value

The median input price across the 443 priced models is $0.482 per million tokens, so Mistral Large 4 Preview (high) costs more than the typical listing. Models that score within 1.5 points or better and cost less: GLM-5.2 (high) ($0.0070, 51.6), DeepSeek V4.1 Flash (high) ($0.010, 49.7), Qwen2.5-Coder-7B-Instruct ($0.010, 49.5). Similar scores cost more on Gemini 3.6 Flash (High) ($0.750) and qwq-plus ($0.800).

Reading these numbers

The 13 scores on this page come from 1 source: Epoch AI Benchmarking Hub (13). No official leaderboard supplied any of them, so treat them as third-party measurements. The leaderboard also lists 2 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

Mistral Large 4 Preview (high) is also listed at 2 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

More from Mistral AI

Compare it on the live leaderboard → · How the aggregate score works