Ranked by GPQA Diamond - 20 models

Best AI models for reasoning

In board data through October 15, 2026, GPT-6 Astra ranks first on GPQA Diamond (vendor-reported) with 96.0%, 1.4 points ahead of GPT-5.6 Sol at 94.6%. Its list price is $10 input / $50 output per 1M tokens. A fixed rule, the lowest output price per 1M tokens among models with a registry price that score within 5 points of the top score, picks GPT-5.6 Luna at 92.3% ($0.20 input / $1.20 output per 1M tokens). The highest-placed open-weight model is DeepSeek V4 Pro, ranked 11th with 92.4.

Board data through October 15, 2026, checked October 17, 2026. How this is ranked

GPQA Diamond ranking

Best AI models for reasoning in GPQA Diamond order, with each score's setting and source, scores on related boards, API price per 1M tokens, context window and weights.
Rank	Model	Provider	GPQA Diamond	Source	Input / 1M	Output / 1M	Context	Weights	Humanity's Last Examthru Oct 17, 2026	ARC-AGI-2thru Oct 13, 2026
1	GPT-6 Astra	OpenAI	96.0%Vendor-reported - OpenAI research environment or API; 'Evaluation scores are the maximum at any effort.'	openai.com	$10	$50	1.05M	Closed	54.80±1.94	95.0%
2	GPT-5.6 Sol	OpenAI	94.6%Vendor-reported - OpenAI GPT-5.6 launch eval table (same value repeated in the GPT-6 Astra table)	openai.com	$4	$20	1.05M	Closed	-	92.5%
3	Claude Mythos Preview	-	94.55%Vendor-reported - Anthropic system card: 'averaged over 5 trials'	www-cdn.anthropic.com	-	-	-	-	-	-
4	Gemini 3.1 Pro PreviewBoard: Gemini 3.1 Pro	Google	94.3%Vendor-reported - Gemini 3.1 Pro Thinking (High), No tools	deepmind.google	$2	$12	1.05M	Closed	46.44±1.96	-
5	Claude Opus 4.7	Anthropic	94.2%Vendor-reported - Anthropic system card: 'averaged over 10 trials'	www-cdn.anthropic.com	$5	$25	1M	Closed	36.20±1.88	-
6	Claude Mythos 5	Anthropic	94.1%Vendor-reported - Claude Fable 5 & Claude Mythos 5 System Card section 8.8: 'averaged over 5 trials'; Anthropic notes it considers GPQA Diamond saturated and plans to stop reporting it (same 94.1% shown for Claude Mythos 5 in OpenAI's GPT-5.6 table)	www-cdn.anthropic.com	$10	$50	1M	Closed	-	-
7	GPT-5.5	OpenAI	93.6%Vendor-reported - 'Evals of GPT were run with reasoning effort set to xhigh' in a research environment	openai.com	$5	$30	1.05M	Closed	-	-
7	Claude Opus 4.8	Anthropic	93.6%Vendor-reported - Anthropic system card: 'averaged over 25 trials'	www-cdn.anthropic.com	$5	$25	1M	Closed	-	-
9	GPT-5.6 Terra	OpenAI	92.9%Vendor-reported - OpenAI GPT-5.6 launch eval table	openai.com	$2	$12	1.05M	Closed	-	-
10	Qwen3.8-Max	Alibaba Cloud (Qwen)	92.6Vendor-reported - Qwen3.8-Max full benchmark table (weights published as Qwen/Qwen3.8-2.4T-A95B)	qwen.ai	$2	$6	1M	Closed	-	-
11	DeepSeek V4 ProBoard: DeepSeek-V4-Pro	DeepSeek	92.4Vendor-reported - GPQA Diamond (Pass@1), 'Comparison with frontier models (Max reasoning effort)' table in DeepSeek-V4.1-Flash model card (original V4-Pro card lists DS-V4-Pro Max at 90.1)	huggingface.co	$1.32	$3.96	1M	Open	-	-
12	GPT-5.6 Luna	OpenAI	92.3%Vendor-reported - OpenAI GPT-5.6 launch eval table	openai.com	$0.20	$1.20	1.05M	Closed	-	-
13	Gemini 3 Pro	-	91.9%Vendor-reported - Gemini 3 Pro Thinking (High), No tools (as listed in the Gemini 3.1 Pro model card and the Gemini 3 launch post)	blog.google	-	-	-	-	-	-
14	Qwen3.8-Flash-Next	-	91.7Vendor-reported - Qwen3.8-Flash-Next Language benchmark table (Scientific reasoning)	qwen.ai	-	-	-	-	-	-
15	GLM-5.2	Z.ai	91.2Vendor-reported - Z.ai full benchmark table; reasoning tasks use temperature=1.0, top_p=0.95, max generation length 163,840 tokens; open weights	z.ai	$1.40	$4.40	1M	Open	-	-
16	DeepSeek V4.1 FlashBoard: DeepSeek-V4.1-Flash	DeepSeek	90.9Vendor-reported - GPQA Diamond (Pass@1), Max reasoning effort; open weights	huggingface.co	$0.30	$1.20	1M	Open	-	-
17	Gemini 3 Flash PreviewBoard: Gemini 3 Flash	Google	90.4%Vendor-reported - Gemini 3 Flash launch post	blog.google	$0.50	$3	1.05M	Closed	-	-
18	Qwen3.8-27B	Alibaba Cloud (Qwen)	89.2Vendor-reported - Qwen model card Text Performance table (Scientific reasoning); open weights	huggingface.co	$0.50	$3	1M	Open	-	-
19	Muse Spark Open (816B-A42B)	-	89.0Vendor-reported - Muse Spark Open (xhigh): 'xhigh reasoning effort, temperature = 1.0, top_p=1.0, top_k=disabled across benchmarks'; open weights	huggingface.co	-	-	-	-	-	-
20	Gemma 4 31B	-	84.3%Vendor-reported - Gemma 4 model card benchmark results; open weights	ai.google.dev	-	-	-	-	-	-

Prices in USD per 1M tokens at the standard rate, from each provider's pricing page. A dash means the board name does not match a model in our registry or the model is not on that board. Each related-board score links to its source.

How this is ranked

Board: GPQA Diamond
What it measures: GPQA is a dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry, and GPQA Diamond is its 198-question subset where both expert validators answered correctly and at most one of three non-experts did.
Data through: October 15, 2026 - newest entry date on the board, checked October 17, 2026. Pages publish only while it is within 120 days of the check.
Rows ranked: 20 models (20 vendor-reported)

Order is the board's own, never a Developers Digest score.

Settings and reporters differ between rows: official-leaderboard scores come from the board's own runs, vendor-reported scores from the model maker's card under its own settings, and agents, effort levels and harnesses vary. A single benchmark is not a verdict on a model.

Board notes (10)

No official GPQA Diamond leaderboard exists (the benchmark authors publish only the paper arXiv:2311.12022 and the github.com/idavidrein/gpqa repo); all entries are vendor-reported from model cards and release posts. Third-party aggregator leaderboards were excluded.
Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, Claude Fable 5.1 (self-reported): Anthropic launch-page benchmark grids omit GPQA Diamond, and the Fable 5 & Mythos 5 system card says Anthropic considers GPQA Diamond saturated and plans to stop reporting it. Fable 5.1, Opus 5 and Fable 5 values come from OpenAI comparison tables instead.
GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna: their OpenAI launch posts have no GPQA Diamond row.
Gemini 3.8 Flash and Gemini 3.7 Flash: Google's own model card (published 2 September 2026) does not list GPQA Diamond; the 95.3% figure is from OpenAI's GPT-6 Astra table.
GLM-5.3: the zai-org/GLM-5.3 model card benchmark table has no GPQA Diamond row (Meta's Muse Spark Open card lists GLM 5.3 (max) at 91.7, DeepSeek's card lists 88.1; neither is Z.ai's own figure).
MiniMax-M3: the official MiniMaxAI/MiniMax-M3 model card could not be read; only third-party figures were found (Z.ai lists 93.0).
xAI Grok 4.5 / 4.6: no official xAI GPQA Diamond figure found; the latest on x.ai is Grok 4 at 87.5% (pass@1), which is not current.
gpt-oss-120b / gpt-oss-20b: the GPQA Diamond values on the OpenAI gpt-oss post could not be extracted cleanly.
Release dates could not be verified for GPT-6 Astra (post has later update notes dated September 22 and 29, 2026), Claude Mythos Preview, Claude Opus 4.7, Claude Fable 5.1 in OpenAI's table, and Gemma 4 model card; date left blank.
Gemini 3 Pro (2025-11-18) and Gemini 3 Flash (2025-12-17) dates are launch dates as recalled, not re-read from the pages today; Muse Spark Open date (2026-10-15) came from a page summary and should be spot-checked.

Rankings

API capabilities

Reasoning

Providers of the top 3

OpenAI models

FAQ

What is the best AI model for reasoning?

GPT-6 Astra ranks first by GPQA Diamond, board data through October 15, 2026, with 96.0% (vendor-reported). One benchmark is not a verdict: settings differ between rows, so test on your own workload before you switch.

What is the cheapest AI model for reasoning near the top of GPQA Diamond?

Taking every ranked model within 5 points of the top score that has a price in our registry, GPT-5.6 Luna has the lowest output price: $0.20 input / $1.20 output per 1M tokens, with 92.3% by GPQA Diamond, board data through October 15, 2026.

What is the best open-weight model for reasoning?

DeepSeek V4 Pro is the highest-placed open-weight model by GPQA Diamond, board data through October 15, 2026, ranked 11th with 92.4.

How is this reasoning ranking built?

Rows follow the GPQA Diamond board's own order, checked on October 17, 2026; its newest entry is dated October 15, 2026, and a use case is only published while that date is within 120 days of the check. Each model appears once. Prices, context windows and weights come from each provider's first-party pricing page in our model registry; a model the registry does not list shows a dash.

Newsletter

One email per week. New model releases, the leaderboard moves that matter and what they mean for what you build.