GPT-5 mini (high): benchmark scores, price and ranking

GPT-5 mini (high) from OpenAI ranks #478 of 743 LLMs on the LLMs Tiger aggregate benchmark score (46.9), based on 15 benchmark results captured up to 2026-10-13.

Organisation

OpenAI

Type

Fast / low-cost (flash) model

Open weights

No

Released

2025-08-07 (source: curated)

Cheapest API price

$0.250 in / $2.00 out per million tokens (via azure, observed 2026-10-13T19:47Z)

Aggregate rank

#478 of 743 (score 46.9)

Relative to other models, GPT-5 mini (high) is strongest on Chess Puzzles (#26 of 122) and weakest on Mystery Games (#55 of 59).

Across the 15 benchmarks where at least ten models are scored, GPT-5 mini (high) is in the top 10% on 0 and in the top half on 7.

Its cheapest listed input price, $0.250 per million tokens, is lower than that of 64% of the 427 models with a listed price.

GPT-5 mini (high) benchmark results

About GPT-5 mini (high)

GPT-5 mini (high) is a smaller, lower-cost model from OpenAI. It was released on 2025-08-07, according to the developer's announcement. The cheapest API listing we track charges $0.250 per million input tokens and $2.00 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 GPT-5 mini (high) stands

On our aggregate score, GPT-5 mini (high) is #478 of 743 ranked models, in the lower half. Its score is 46.9; the scale is centred on 50, so it sits 3.1 points below the typical model. That beats 36% of the ranked field. Among the 88 ranked OpenAI models it is #65. The next one up is GPT-4 Turbo Preview (January 2024) at 46.9. The next one down is GPT-4 (1106) at 46.9.

GPT-5 mini (high), benchmark by benchmark

Chess Puzzles puts GPT-5 mini (high) at 30.0%, #26 of 122. That beats 79% of the models tested, but 42.0 points short of GPT-6 Astra (max) (72.0%). The results either side are Grok 4.6 (xhigh) (31.0%) and Qwen3.8 Max (xhigh) (29.0%). It covers chess tactics puzzles. The board median is 12.5%, so GPT-5 mini (high) is 17.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On WeirdML, GPT-5 mini (high) reaches 52.7%. It ranks #25 of 109, ahead of 77% of the field, and 40.6 points behind GPT-6 Astra (max) at 93.3%. The results either side are gemini-2.5-pro_16K (54.0%) and o4-mini (high) (52.6%). It covers unusual machine-learning tasks. The board median is 41.2%, so GPT-5 mini (high) is 11.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 mini (high) manages 86.7% on OTIS Mock AIME, good for #51 of 189. Claude Fable 5.1 (max) leads at 100.0%, 13.3 points further up, and 73% of models sit below. The results either side are Claude Opus 4.7 (max) (86.7%) and nemotron-3-ultra (86.7%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so GPT-5 mini (high) is 28.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

DTBench puts GPT-5 mini (high) at 80.5%, #50 of 146. That beats 66% of the models tested, but 18.4 points short of Claude Opus 5.5 (max) (98.9%). The results either side are qwen-plus-2025-04-28 (81.1%) and Qwen3.5 Plus (80.5%). It covers dynamic tool-use benchmark. The board median is 68.8%, so GPT-5 mini (high) is 11.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On LMCA, GPT-5 mini (high) reaches 34.2%. It ranks #45 of 108, ahead of 58% of the field, and 34.0 points behind Claude Opus 5.5 (max) at 68.2%. The results either side are Qwen3.6 27B (34.5%) and Qwen3.5 27B (34.0%). It covers language-model capability assessment. The board median is 29.2%, so GPT-5 mini (high) is 5.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 mini (high) manages 800 on ALE-Bench, good for #35 of 79. GPT-6 Astra (max) leads at 2,951, 2,152 points further up, and 56% of models sit below. The results either side are DeepSeek-R1 (May 2025) (804) and Mercury 2 (786). It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 764, so GPT-5 mini (high) is 36 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 mini (high) lands below the middle on SimpleQA Verified: 21.6%, #55 of 69, with 78% of the field scoring higher. gemini-4-argon-high leads at 76.0%; the closest result above is Claude Haiku 5.5 (max) at 23.8%. It covers short factual questions. The board median is 41.0%, so GPT-5 mini (high) is 19.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

FrontierMath T4 is a weak spot. GPT-5 mini (high) scores 12.2%, #41 of 50, behind 80% 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 GPT-5 mini (high) is 18.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

CritPt is a weak spot. GPT-5 mini (high) scores 0.0%, #96 of 115, behind 83% of the models measured. The leader, GPT-5.6 Sol (max), reaches 32.3%. It covers unpublished research-level physics problems. The board median is 2.9%, so GPT-5 mini (high) is 2.9 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Mystery Games is a weak spot. GPT-5 mini (high) scores 5.0%, #55 of 59, behind 92% 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 GPT-5 mini (high) is 18.0 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: uneven, ahead of 45% of the field on average across 2 benchmarks. Its best board here is ALE-Bench (#35 of 79) and its weakest is SciCode (#70 of 106). Reasoning, maths and science: uneven, ahead of 36% of the field on average across 7 benchmarks. Its best board here is OTIS Mock AIME (#51 of 189) and its weakest is CritPt (#96 of 115). General knowledge and chat: uneven, ahead of 48% of the field on average across 5 benchmarks. Its best board here is Chess Puzzles (#26 of 122) and its weakest is Mystery Games (#55 of 59).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so GPT-5 mini (high) costs less 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 DeepSeek-V3.2 (Thinking; Fireworks) ($0.259) and DeepSeek-V3.2 ($0.259).

Release history

The OpenAI model released before it was GPT-5 (high) on 2025-08-07, which scores 51.1. The next one was GPT-5 nano (high) on 2025-08-07, at 43.8.

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 5 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

GPT-5 mini (high) is also listed at 5 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

• #475 Qwen3-32B (46.9)
• #476 GPT-4 Turbo Preview (Nov 2023) (46.9)
• #477 GPT-4 Turbo Preview (January 2024) (46.9)
• #479 GPT-4 (1106) (46.9)
• #480 glm-4-9b (46.9)
• #481 Llama-4-Maverick-17B-128E-Instruct (46.9)

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