Grok 4.5 (high): benchmark scores, price and ranking

Grok 4.5 (high) from xAI ranks #85 of 743 LLMs on the LLMs Tiger aggregate benchmark score (52.0), based on 19 benchmark results captured up to 2026-10-13.

Organisation

xAI

Type

Frontier model

Open weights

No

Released

2026-07-08 (source: epoch)

Cheapest API price

$2.00 in / $6.00 out per million tokens (via openrouter, observed 2026-10-13T19:47Z)

Aggregate rank

#85 of 743 (score 52.0)

Relative to other models, Grok 4.5 (high) is strongest on GPQA Diamond (#9 of 275) and weakest on DeepSWE (#17 of 19).

Across the 17 benchmarks where at least ten models are scored, Grok 4.5 (high) is in the top 10% on 2 and in the top half on 13.

Its cheapest listed input price, $2.00 per million tokens, is lower than that of 18% of the 427 models with a listed price.

Grok 4.5 (high) benchmark results

About Grok 4.5 (high)

Grok 4.5 (high) is a frontier model from xAI. It was released on 2026-07-08, according to Epoch AI. The cheapest API listing we track charges $2.00 per million input tokens and $6.00 per million output tokens, through openrouter. We hold 19 benchmark results for it across 4 areas, and 2 are in the top tenth of their boards.

Where Grok 4.5 (high) stands

On our aggregate score, Grok 4.5 (high) is #85 of 743 ranked models, in the top quarter. Its score is 52.0; the scale is centred on 50, so it sits 2.0 points above the typical model. That beats 89% of the ranked field. Among the 26 ranked xAI models it is #5. The next one up is Grok 4.7 (xhigh) at 53.1. The next one down is Grok 4.20 (Reasoning) at 51.2.

Grok 4.5 (high), benchmark by benchmark

On GPQA Diamond, Grok 4.5 (high) scores 93.4% and ranks #9 of 275. Only 8 models score higher, led by GPT-6 Astra (max) at 95.8%, which is 2.3 points ahead. It covers expert-level science questions. The board median is 71.7%, so Grok 4.5 (high) is 21.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On DTBench, Grok 4.5 (high) scores 96.5% and ranks #7 of 146. Only 6 models score higher, led by Claude Opus 5.5 (max) at 98.9%, which is 2.4 points ahead. It covers dynamic tool-use benchmark. The board median is 68.8%, so Grok 4.5 (high) is 27.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Grok 4.5 (high) manages 97.8% on OTIS Mock AIME, good for #20 of 189. Claude Fable 5.1 (max) leads at 100.0%, 2.2 points further up, and 89% of models sit below. The results either side are Grok 4.7 (xhigh) (98.1%) and Gemini 3.7 Flash (high) (97.2%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so Grok 4.5 (high) is 40.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

ALE-Bench puts Grok 4.5 (high) at 1,309, #11 of 79. That beats 86% of the models tested, but 1,642 points short of GPT-6 Astra (max) (2,951). The results either side are DeepSeek V4 Pro 0813 (max) (1,403) and DeepSeek V4 Flash 0731 (max) (1,306). It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 764, so Grok 4.5 (high) is 545 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On Chess Puzzles, Grok 4.5 (high) reaches 36.0%. It ranks #19 of 122, ahead of 84% of the field, and 36.0 points behind GPT-6 Astra (max) at 72.0%. The results either side are GPT-5 (high) (37.0%) and o3 (high) (34.0%). It covers chess tactics puzzles. The board median is 12.5%, so Grok 4.5 (high) is 23.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Grok 4.5 (high) manages 45.2% on LMCA, good for #20 of 108. Claude Opus 5.5 (max) leads at 68.2%, 23.0 points further up, and 81% of models sit below. The results either side are Qwen3.8 Max (xhigh) (46.2%) and GPT-6 Luna (max) (44.5%). It covers language-model capability assessment. The board median is 29.2%, so Grok 4.5 (high) is 16.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Grok 4.5 (high) lands below the middle on FrontierMath T4: 24.4%, #33 of 50, with 64% of the field scoring higher. GPT-6.1 Sol (max) leads at 100.0%; the closest result above is DeepSeek V4 Flash 0731 (max) at 24.4%. It covers research-level maths, tier 4. The board median is 30.5%, so Grok 4.5 (high) is 6.1 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GDP.pdf is a weak spot. Grok 4.5 (high) scores 14.0%, #22 of 25, behind 84% of the models measured. The leader, GPT-6 Astra (max), reaches 34.2%. It covers economically valuable document tasks. The board median is 22.8%, so Grok 4.5 (high) is 8.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Furniture Assembly is a weak spot. Grok 4.5 (high) scores 22.5%, #25 of 28, behind 86% of the models measured. The leader, Claude Opus 5.5 (max), reaches 83.3%. It covers planning a physical assembly. The board median is 41.2%, so Grok 4.5 (high) is 18.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

DeepSWE is a weak spot. Grok 4.5 (high) scores 53.8%, #17 of 19, behind 84% of the models measured. The leader, Gemini 3.8 Flash (high), reaches 73.8%. It covers long-horizon software engineering. The board median is 67.0%, so Grok 4.5 (high) is 13.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 9 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: solid, ahead of 58% of the field on average across 3 benchmarks. Its best board here is ALE-Bench (#11 of 79) and its weakest is DeepSWE (#17 of 19). Reasoning, maths and science: solid, ahead of 66% of the field on average across 7 benchmarks. Its best board here is GPQA Diamond (#9 of 275) and its weakest is FrontierMath T4 (#33 of 50). Agents and tool use: solid, ahead of 54% of the field on average across 2 benchmarks. Its best board here is DTBench (#7 of 146) and its weakest is GDP.pdf (#22 of 25). General knowledge and chat: solid, ahead of 64% of the field on average across 5 benchmarks. Its best board here is Chess Puzzles (#19 of 122) and its weakest is Furniture Assembly (#25 of 28).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so Grok 4.5 (high) costs more than the typical listing. Models that score within 1.5 points or better and cost less: GLM-5.2 (high) ($0.060, 51.6), DeepSeek V4 Flash 0731 (max) ($0.070, 53.1), Qwen3-Coder-30B-A3B-Instruct ($0.070, 52.7). Similar scores cost more on Grok 4 ($3.00) and claude-3.5-sonnet-20240620 ($3.00).

Release history

The xAI model released before it was grok-build-0.1 on 2026-05-29, which scores 45.4. The next one was Grok 4.6 (xhigh) on 2026-08-12, at 53.6.

Reading these numbers

The 19 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (17), ARC Prize official leaderboard (1), SWE-rebench leaderboard (1). 1 of them come from an official leaderboard and 18 from datasets or third-party runs. The leaderboard also lists 4 other reasoning-effort settings of this model, which redirect here. This page shows the highest.

Grok 4.5 (high) is also listed at 4 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

• #82 Claude Haiku 5.5 (max) (52.1)
• #83 NVIDIA-Nemotron-3-Super-120B-A12B-BF16 (52.0)
• #84 K-EXAONE-236B-A23B (52.0)
• #86 Darwin-398B-JGOS (52.0)
• #87 SWE-2 (max) (51.9)
• #88 Hy3 (51.9)

More from xAI

• Grok 4.6 (xhigh) (#34, 53.6)
• grok-4-0709 (#38, 53.2)
• Grok 4 (#39, 53.2)
• Grok 4.7 (xhigh) (#41, 53.1)
• Grok 4.20 (Reasoning) (#136, 51.2)
• grok-4-1-fast-non-reasoning (#148, 51.0)

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