o3-mini (high): benchmark scores, price and ranking

o3-mini (high) from OpenAI ranks #649 of 743 LLMs on the LLMs Tiger aggregate benchmark score (44.0), based on 15 benchmark results captured up to 2026-10-13.

Organisation

OpenAI

Type

Fast / low-cost (flash) model

Open weights

No

Released

2025-01-31 (source: curated)

Cheapest API price

$1.10 in / $4.40 out per million tokens (via azure, observed 2026-10-13T19:47Z)

Aggregate rank

#649 of 743 (score 44.0)

Relative to other models, o3-mini (high) is strongest on OTIS Mock AIME (#70 of 189) and weakest on FrontierMath T4 (#50 of 50).

Across the 15 benchmarks where at least ten models are scored, o3-mini (high) is in the top 10% on 0 and in the top half on 4.

Its cheapest listed input price, $1.10 per million tokens, is lower than that of 33% of the 427 models with a listed price.

o3-mini (high) benchmark results

About o3-mini (high)

o3-mini (high) is a smaller, lower-cost model from OpenAI. It was released on 2025-01-31, according to the developer's announcement. The cheapest API listing we track charges $1.10 per million input tokens and $4.40 per million output tokens, through azure. We hold 15 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where o3-mini (high) stands

On our aggregate score, o3-mini (high) is #649 of 743 ranked models, in the lower half. Its score is 44.0; the scale is centred on 50, so it sits 6.0 points below the typical model. That beats 13% of the ranked field. Among the 88 ranked OpenAI models it is #77. The next one up is gpt-oss-20b (high) at 44.3. The next one down is GPT-5 nano (high) at 43.8.

o3-mini (high), benchmark by benchmark

OTIS Mock AIME puts o3-mini (high) at 76.9%, #70 of 189. That beats 63% of the models tested, but 23.1 points short of Claude Fable 5.1 (max) (100.0%). The results either side are grok-3-mini-beta (high) (77.8%) and Gemma 4 31B IT (73.3%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so o3-mini (high) is 19.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On WeirdML, o3-mini (high) reaches 43.7%. It ranks #46 of 109, ahead of 58% of the field, and 49.6 points behind GPT-6 Astra (max) at 93.3%. The results either side are Mistral Medium 3.5 (43.7%) and nemotron-3-ultra (43.5%). It covers unusual machine-learning tasks. The board median is 41.2%, so o3-mini (high) is 2.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

o3-mini (high) manages 77.0% on GPQA Diamond, good for #117 of 275. GPT-6 Astra (max) leads at 95.8%, 18.8 points further up, and 57% of models sit below. The results either side are NVIDIA-Nemotron-3-Super-120B-A12B-FP8 (77.3%) and claude-opus-4-1-20250805_27K (76.8%). It covers expert-level science questions. The board median is 71.7%, so o3-mini (high) is 5.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Chess Puzzles puts o3-mini (high) at 17.0%, #52 of 122. That beats 57% of the models tested, but 55.0 points short of GPT-6 Astra (max) (72.0%). The results either side are Inkling Small (xhigh) (18.0%) and Qwen 3.6 Plus (2026-04-02) (17.0%). It covers chess tactics puzzles. The board median is 12.5%, so o3-mini (high) is 4.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

o3-mini (high) lands below the middle on DTBench: 68.8%, #74 of 146, with 50% of the field scoring higher. Claude Opus 5.5 (max) leads at 98.9%; the closest result above is GPT-4.1 mini at 68.8%. It covers dynamic tool-use benchmark. The median score on this board is also 68.8%. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

o3-mini (high) lands below the middle on SciCode: 39.8%, #68 of 106, with 63% of the field scoring higher. Claude Opus 5.5 (max) leads at 66.9%; the closest result above is Step 3.7 Flash at 40.0%. It covers scientific research code. The board median is 44.3%, so o3-mini (high) is 4.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Mystery Games is a weak spot. o3-mini (high) scores 7.0%, #52 of 59, behind 86% of the models measured. The leader, GPT-6 Astra (max), reaches 84.0%. It covers deductive reasoning puzzles. The board median is 23.0%, so o3-mini (high) is 16.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

SimpleQA Verified is a weak spot. o3-mini (high) scores 15.3%, #61 of 69, behind 87% of the models measured. The leader, gemini-4-argon-high, reaches 76.0%. It covers short factual questions. The board median is 41.0%, so o3-mini (high) is 25.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GSO-Bench is a weak spot. o3-mini (high) scores 1.3%, #17 of 19, behind 84% of the models measured. The leader, GPT-5.5 (xhigh), reaches 40.2%. It covers software performance optimisation. The board median is 4.9%, so o3-mini (high) is 3.6 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

FrontierMath T4 is a weak spot. o3-mini (high) scores 0.0%, #50 of 50, behind 98% of the models measured. The leader, GPT-6.1 Sol (max), reaches 100.0%. It covers research-level maths, tier 4. The board median is 30.5%, so o3-mini (high) is 30.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 5 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: weak, ahead of 23% of the field on average across 2 benchmarks. Its best board here is SciCode (#68 of 106) and its weakest is GSO-Bench (#17 of 19). Reasoning, maths and science: weak, ahead of 29% of the field on average across 7 benchmarks. Its best board here is OTIS Mock AIME (#70 of 189) and its weakest is FrontierMath T4 (#50 of 50). General knowledge and chat: uneven, ahead of 34% of the field on average across 5 benchmarks. Its best board here is WeirdML (#46 of 109) and its weakest is SimpleQA Verified (#61 of 69).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so o3-mini (high) costs more than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on Mistral Medium 3.5 ($1.50) and GPT-4.1 ($2.00).

Release history

The OpenAI model released before it was o3-mini-2025-01-31 (low) on 2025-01-31, which scores 46.0. The next one was chatgpt-4o-latest (2025-02-15) on 2025-02-15, at 48.6.

Reading these numbers

The 15 scores on this page come from 2 sources: Epoch AI Benchmarking Hub (14), ARC Prize official leaderboard (1). 1 of them come from an official leaderboard and 14 from datasets or third-party runs. The leaderboard also lists 3 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

o3-mini (high) is also listed at 3 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

• #646 Claude 3.5 Sonnet (Jun 2024) (44.2)
• #647 DeepSeek-V3 (Mar 2025) (44.1)
• #648 Yi-34B-Chat (44.0)
• #650 gemini-1.5-pro-001 (44.0)
• #651 Mellum2-12B-A2.5B-Base (44.0)
• #652 Mellum2-12B-A2.5B-Base-Pretrain (44.0)

More from OpenAI

• GPT-6 Astra (max) (#2, 64.0)
• GPT-6.1 Sol (max) (#5, 60.6)
• GPT-5.6 Sol (max) (#6, 60.1)
• GPT-6 Sol (max) (#7, 60.0)
• GPT-5.5 (xhigh) (#11, 58.0)
• GPT-5.6 Terra (max) (#12, 57.8)

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