GPT-5.2 (xhigh): benchmark scores, price and ranking

GPT-5.2 (xhigh) from OpenAI ranks #90 of 743 LLMs on the LLMs Tiger aggregate benchmark score (51.9), based on 14 benchmark results captured up to 2026-10-13.

Organisation

OpenAI

Type

Frontier model

Open weights

No

Released

2025-12-11 (source: curated)

Cheapest API price

$1.75 in / $14.00 out per million tokens (via azure, observed 2026-10-13T19:47Z)

Aggregate rank

#90 of 743 (score 51.9)

Relative to other models, GPT-5.2 (xhigh) is strongest on Chess Puzzles (#6 of 122) and weakest on ProofBench (#32 of 46).

Across the 14 benchmarks where at least ten models are scored, GPT-5.2 (xhigh) is in the top 10% on 2 and in the top half on 9.

Its cheapest listed input price, $1.75 per million tokens, is lower than that of 24% of the 427 models with a listed price.

GPT-5.2 (xhigh) benchmark results

About GPT-5.2 (xhigh)

GPT-5.2 (xhigh) is a frontier model from OpenAI. It was released on 2025-12-11, according to the developer's announcement. The cheapest API listing we track charges $1.75 per million input tokens and $14.00 per million output tokens, through azure. We hold 14 benchmark results for it across 3 areas, and 2 are in the top tenth of their boards.

Where GPT-5.2 (xhigh) stands

On our aggregate score, GPT-5.2 (xhigh) is #90 of 743 ranked models, in the top quarter. Its score is 51.9; the scale is centred on 50, so it sits 1.9 points above the typical model. That beats 88% of the ranked field. Among the 88 ranked OpenAI models it is #14. The next one up is gpt-5.2-2025-12-11-high at 52.1. The next one down is gpt-4o-2024-05-13 at 51.9.

GPT-5.2 (xhigh), benchmark by benchmark

On Chess Puzzles, GPT-5.2 (xhigh) scores 49.0% and ranks #6 of 122. Only 5 models score higher, led by GPT-6 Astra (max) at 72.0%, which is 23.0 points ahead. It covers chess tactics puzzles. The board median is 12.5%, so GPT-5.2 (xhigh) is 36.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On GPQA Diamond, GPT-5.2 (xhigh) scores 91.4% and ranks #25 of 275. Only 24 models score higher, led by GPT-6 Astra (max) at 95.8%, which is 4.4 points ahead. It covers expert-level science questions. The board median is 71.7%, so GPT-5.2 (xhigh) is 19.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5.2 (xhigh) manages 72.2% on WeirdML, good for #13 of 109. GPT-6 Astra (max) leads at 93.3%, 21.1 points further up, and 88% of models sit below. The results either side are GLM-5.3 (max) (75.4%) and Gemini 3 Pro Preview (69.9%). It covers unusual machine-learning tasks. The board median is 41.2%, so GPT-5.2 (xhigh) is 31.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

OTIS Mock AIME puts GPT-5.2 (xhigh) at 96.1%, #25 of 189. That beats 87% of the models tested, but 3.9 points short of Claude Fable 5.1 (max) (100.0%). The results either side are Kimi K2.6 (96.1%) and Kimi K2.7 Code (95.6%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so GPT-5.2 (xhigh) is 38.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On DTBench, GPT-5.2 (xhigh) reaches 90.9%. It ranks #21 of 146, ahead of 86% of the field, and 8.0 points behind Claude Opus 5.5 (max) at 98.9%. The results either side are DeepSeek V4 Flash 0731 (max) (90.9%) and Kimi K2.6 (90.9%). It covers dynamic tool-use benchmark. The board median is 68.8%, so GPT-5.2 (xhigh) is 22.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5.2 (xhigh) manages 43.9% on LMCA, good for #23 of 108. Claude Opus 5.5 (max) leads at 68.2%, 24.3 points further up, and 79% of models sit below. The results either side are Qwen3.7 Max (44.0%) and GPT-5.1 (high) (43.9%). It covers language-model capability assessment. The board median is 29.2%, so GPT-5.2 (xhigh) is 14.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5.2 (xhigh) lands below the middle on Furniture Assembly: 38.3%, #16 of 28, with 54% of the field scoring higher. Claude Opus 5.5 (max) leads at 83.3%; the closest result above is Grok 4.6 (xhigh) at 40.0%. It covers planning a physical assembly. The board median is 41.2%, so GPT-5.2 (xhigh) is 2.9 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5.2 (xhigh) lands below the middle on MCP Atlas: 67.6%, #9 of 14, with 57% of the field scoring higher. Muse Spark leads at 82.2%; the closest result above is Gemini 3 Pro Preview at 70.3%. It covers tool use across 1,000 MCP tasks. The board median is 70.4%, so GPT-5.2 (xhigh) is 2.8 points below it. Source: Scale Labs MCP Atlas leaderboard, captured 2026-10-13.

GPT-5.2 (xhigh) lands below the middle on EBR-Bench: 23.0%, #14 of 21, with 62% of the field scoring higher. GPT-6 Astra (max) leads at 76.2%; the closest result above is GPT-5.4 (xhigh) at 25.4%. It covers evidence-based reasoning. The board median is 31.4%, so GPT-5.2 (xhigh) is 8.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPT-5.2 (xhigh) lands below the middle on ProofBench: 15.0%, #32 of 46, with 67% of the field scoring higher. Claude Fable 5.1 (max) leads at 100.0%; the closest result above is mimo-v2.5 at 16.0%. It covers writing mathematical proofs. The board median is 21.0%, so GPT-5.2 (xhigh) is 6.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 4 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Reasoning, maths and science: solid, ahead of 64% of the field on average across 6 benchmarks. Its best board here is GPQA Diamond (#25 of 275) and its weakest is ProofBench (#32 of 46). Agents and tool use: solid, ahead of 61% of the field on average across 2 benchmarks. Its best board here is DTBench (#21 of 146) and its weakest is MCP Atlas (#9 of 14). General knowledge and chat: solid, ahead of 64% of the field on average across 6 benchmarks. Its best board here is Chess Puzzles (#6 of 122) and its weakest is EBR-Bench (#14 of 21).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so GPT-5.2 (xhigh) 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), DeepSeek V4 Flash 0731 (max) ($0.070, 53.1). Similar scores cost more on Grok 4.7 (xhigh) ($2.00) and Claude Sonnet 5 (max) ($2.00).

Release history

The OpenAI model released before it was gpt-5.2-2025-12-11-high on 2025-12-11, which scores 52.1. The next one was GPT-5.2 Codex on 2025-12-18, at 52.2.

Reading these numbers

The 14 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (12), ARC Prize official leaderboard (1), Scale Labs MCP Atlas leaderboard (1). 2 of them come from an official leaderboard and 12 from datasets or third-party runs. The leaderboard also lists 6 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

GPT-5.2 (xhigh) is also listed at 6 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

• #87 SWE-2 (max) (51.9)
• #88 Hy3 (51.9)
• #89 Dots3-Note Preview (Max) (51.9)
• #91 Hy4-preview (51.9)
• #92 gpt-4o-2024-05-13 (51.9)
• #93 ollama/qwen2.5-coder:32b (51.9)

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