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

GPT-5 (high) from OpenAI ranks #144 of 743 LLMs on the LLMs Tiger aggregate benchmark score (51.1), based on 21 benchmark results captured up to 2026-10-13.

Organisation

OpenAI

Type

Frontier model

Open weights

No

Released

2025-08-07 (source: curated)

Cheapest API price

$1.25 in / $10.00 out per million tokens (via azure, observed 2026-10-13T19:47Z)

Aggregate rank

#144 of 743 (score 51.1)

Relative to other models, GPT-5 (high) is strongest on Chess Puzzles (#18 of 122) and weakest on EBR-Bench (#18 of 21).

Across the 20 benchmarks where at least ten models are scored, GPT-5 (high) is in the top 10% on 0 and in the top half on 14.

Its cheapest listed input price, $1.25 per million tokens, is lower than that of 26% of the 427 models with a listed price.

GPT-5 (high) benchmark results

About GPT-5 (high)

GPT-5 (high) is a frontier model from OpenAI. It was released on 2025-08-07, according to the developer's announcement. The cheapest API listing we track charges $1.25 per million input tokens and $10.00 per million output tokens, through azure. We hold 21 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where GPT-5 (high) stands

On our aggregate score, GPT-5 (high) is #144 of 743 ranked models, in the top quarter. Its score is 51.1; the scale is centred on 50, so it sits 1.1 points above the typical model. That beats 81% of the ranked field. Among the 88 ranked OpenAI models it is #19. The next one up is o1-2024-12-17 at 51.6. The next one down is GPT-5.6 Luna 2026-07-30 (Max) at 50.9.

GPT-5 (high), benchmark by benchmark

Chess Puzzles puts GPT-5 (high) at 37.0%, #18 of 122. That beats 85% of the models tested, but 35.0 points short of GPT-6 Astra (max) (72.0%). The results either side are Muse Spark 1.3 (max) (38.0%) and Grok 4.5 (high) (36.0%). It covers chess tactics puzzles. The board median is 12.5%, so GPT-5 (high) is 24.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On METR time horizon, GPT-5 (high) reaches 203 min. It ranks #5 of 31, ahead of 84% of the field, and 842 minutes behind Claude Mythos Preview (Early) at 1,045 min. The results either side are GPT-5.1-Codex-Max (224 min) and claude-opus-4-1-20250805_16K (114 min). It covers length of task (minutes) done at 50% success. The board median is 23 min, so GPT-5 (high) is 180 minutes above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 (high) manages 90.7% on DTBench, good for #24 of 146. Claude Opus 5.5 (max) leads at 98.9%, 8.3 points further up, and 84% of models sit below. The results either side are DeepSeek v4 Pro (max) (90.7%) and grok-4.3 (high) (90.7%). It covers dynamic tool-use benchmark. The board median is 68.8%, so GPT-5 (high) is 21.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

SWE-bench Verified puts GPT-5 (high) at 73.6%, #12 of 73. That beats 84% of the models tested, but 9.9 points short of Claude Opus 4.7 (max) (83.5%). The results either side are GLM-5.1 (74.2%) and claude-opus-4-1-20250805 (73.3%). It covers real GitHub issues resolved. The board median is 56.4%, so GPT-5 (high) is 17.2 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On WeirdML, GPT-5 (high) reaches 60.7%. It ranks #18 of 109, ahead of 83% of the field, and 32.6 points behind GPT-6 Astra (max) at 93.3%. The results either side are GPT-5.1 (high) (60.8%) and Muse Spark 1.2 (xhigh) (60.3%). It covers unusual machine-learning tasks. The board median is 41.2%, so GPT-5 (high) is 19.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 (high) manages 25.3% on Humanity's Last Exam, good for #5 of 27. Muse Spark leads at 40.6%, 15.2 points further up, and 81% of models sit below. The results either side are Claude Opus 4.6 (max) (34.4%) and Kimi K2.5 (24.4%). It covers expert-level questions across domains. The board median is 10.7%, so GPT-5 (high) is 14.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 (high) lands below the middle on ProofBench: 18.0%, #26 of 46, with 54% of the field scoring higher. Claude Fable 5.1 (max) leads at 100.0%; the closest result above is Gemini 3 Pro Preview at 20.0%. It covers writing mathematical proofs. The board median is 21.0%, so GPT-5 (high) is 3.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5 (high) lands below the middle on ARC-AGI-2: 9.9%, #50 of 87, with 56% of the field scoring higher. GPT-6 Astra (max) leads at 95.0%; the closest result above is Gemini 3.5 Flash-Lite (High) at 10.3%. It covers abstract reasoning, ARC Prize public eval set. The board median is 22.8%, so GPT-5 (high) is 12.9 points below it. Source: ARC Prize official leaderboard, captured 2026-10-13.

GPT-5 (high) lands below the middle on FrontierMath T4: 22.0%, #35 of 50, with 68% of the field scoring higher. GPT-6.1 Sol (max) leads at 100.0%; the closest result above is Gemini 3.8 Flash (high) at 22.0%. It covers research-level maths, tier 4. The board median is 30.5%, so GPT-5 (high) is 8.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

EBR-Bench is a weak spot. GPT-5 (high) scores 12.7%, #18 of 21, behind 81% of the models measured. The leader, GPT-6 Astra (max), reaches 76.2%. It covers evidence-based reasoning. The board median is 31.4%, so GPT-5 (high) is 18.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 11 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: solid, ahead of 74% of the field on average across 3 benchmarks. Its best board here is SWE-bench Verified (#12 of 73) and its weakest is GSO-Bench (#7 of 19). Reasoning, maths and science: solid, ahead of 61% of the field on average across 9 benchmarks. Its best board here is Humanity's Last Exam (#5 of 27) and its weakest is FrontierMath T4 (#35 of 50). General knowledge and chat: solid, ahead of 66% of the field on average across 7 benchmarks. Its best board here is Chess Puzzles (#18 of 122) and its weakest is EBR-Bench (#18 of 21).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so GPT-5 (high) costs more than the typical listing. Models that score within 1.5 points or better and cost less: Qwen3.7 Flash ($0.030, 50.5), GLM-5.2 (high) ($0.060, 51.6), glm-4.7-flash ($0.060, 50.1). Similar scores cost more on glm-5p1 ($1.40) and gemini-3.5-flash-high ($1.50).

Release history

The OpenAI model released before it was gpt-5-chat on 2025-08-07, which scores 49.2. The next one was GPT-5 mini (high) on 2025-08-07, at 46.9.

Reading these numbers

The 21 scores on this page come from 2 sources: Epoch AI Benchmarking Hub (20), ARC Prize official leaderboard (1). 1 of them come from an official leaderboard and 20 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 (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

• #141 GLM-5.1-FP8 (51.2)
• #142 Llama 3.1-405B (51.2)
• #143 Claude 4.5 Sonnet (51.2)
• #145 Claude Code + Fable 5 (51.1)
• #146 Qwen3.5 397B-A17B (51.1)
• #147 Gemini 2.5 Pro Preview (Jun 2025) (51.0)

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