Direct publisher snapshots

Published AI benchmark snapshots

Compare source-linked results while preserving each publisher’s model labels, configurations, dates and scoring rules. Cross-board ranks are not treated as interchangeable. Only scores stated in the publisher page’s rendered summary are reproduced; missing rows are never filled from model cards, estimates, or mixed settings.

Catalogue review recorded 2026-09-04 · individual claims have separate datesStructured datasetSource Methodology

Publication rule: a score is shown only when the named source page states the model, rank, and score together. These are publisher snapshots, not AIBenchIndex measurements. Open the linked source for the full current interactive board.

GPQA Diamond

Source: OpenRouter

OpenRouter default-routing board

Publisher source

GPQA Diamond is a graduate-level multiple-choice benchmark in biology, physics, and chemistry. OpenRouter runs the same fixed question set across provider endpoints with default routing to compare capability, routing, and the practical cost of solving difficult scientific problems.

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#1	Gemini 3.1 Pro Preview	Google DeepMind	94.4%
#2	GPT-6 Astra	OpenAI	94.4%
#3	Gemini 3.7 Flash	Google DeepMind	94.3%
#4	GPT-5.5	OpenAI	93.8%
#5	GPT-5.6 Sol Pro	OpenAI	93.8%
#6	Grok 4.6	xAI	93.3%
#7	Gemini 3.6 Flash	Google DeepMind	92.8%
#8	Gemini 3.5 Flash	Google DeepMind	92.8%
#9	GPT-5.6 Sol	OpenAI	91.9%
#10	Kimi K3	Moonshot AI	91.5%
#11	Claude Fable 5.1	Anthropic	90.9%
#12	MiniMax M3	MiniMax	90.5%
#13	GPT-5.4	OpenAI	90.3%
#14	GPT-5.6 Luna Pro	OpenAI	90.3%
#15	Claude Opus 4.8	Anthropic	89.4%
#16	Nova Micro 1.0	Amazon	89.1%
#17	GPT-5.6 Terra	OpenAI	89.0%
#18	Claude Opus 4.7	Anthropic	88.9%
#19	DeepSeek V4 Pro	DeepSeek	88.9%
#20	Gemini 3 Flash Preview	Google DeepMind	88.3%
#21	DeepSeek V4 Flash Vision Exp	DeepSeek	88.2%
#22	GPT-5.2	OpenAI	87.9%
#23	GPT-5.6 Luna	OpenAI	87.7%
#24	Claude Opus 5	Anthropic	86.8%
#25	GLM 5.3 Flash	Z.ai	86.7%
#26	DeepSeek V4 Flash 0423	DeepSeek	86.6%
#27	Claude Opus 4.5	Anthropic	86.6%
#28	Inkling Small	Thinking Machines	86.5%
#29	DeepSeek V4 Pro 0423	DeepSeek	86.4%
#30	GPT-5.1	OpenAI	86.2%

Show remaining 99 of 129 rows from the source board

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#31	GPT-5	OpenAI	86.0%
#32	Qwen3.5 397B A17B	Alibaba Cloud	85.9%
#33	GLM 5.2	Z.ai	85.8%
#34	Qwen3.8 2.4T A95B	Alibaba Cloud	85.7%
#35	Claude Opus 4.6	Anthropic	85.6%
#36	DeepSeek V4 Flash 0731	DeepSeek	85.2%
#37	MiniMax M2.1	MiniMax	85.1%
#38	Kimi K2.5	Moonshot AI	84.9%
#39	GLM 5.3	Z.ai	84.8%
#40	MiniMax M2.5	MiniMax	84.2%
#41	MiMo-V2.5-Pro	Xiaomi	84.2%
#42	MiniMax M2.7	MiniMax	84.1%
#43	GPT-5.4 Mini	OpenAI	83.7%
#44	Kimi K2 Thinking	Moonshot AI	83.6%
#45	Qwen3.5-122B-A10B	Alibaba Cloud	83.6%
#46	Gemini 3.5 Flash Lite	Google DeepMind	83.5%
#47	GLM 4.7	Z.ai	83.5%
#48	Gemma 4 31B	Google DeepMind	83.4%
#49	Kimi K2.6	Moonshot AI	83.3%
#50	Inkling	Thinking Machines	83.2%
#51	GLM 5.1	Z.ai	82.9%
#52	Claude Fable 5	Anthropic	82.8%
#53	Claude Sonnet 4.5	Anthropic	82.6%
#54	Claude Sonnet 5	Anthropic	82.2%
#55	Nemotron 3 Ultra	NVIDIA	82.2%
#56	DeepSeek V3.2	DeepSeek	82.0%
#57	Qwen3.8 27B	Alibaba Cloud	81.8%
#58	Qwen3.5-35B-A3B	Alibaba Cloud	81.7%
#59	Kimi K2.7 Code	Moonshot AI	80.7%
#60	MiMo-V2-Flash	Xiaomi	80.7%
#61	Qwen3.6 35B A3B	Alibaba Cloud	80.6%
#62	GPT-5.3 Chat	OpenAI	80.5%
#63	DeepSeek V3.1 Terminus	DeepSeek	80.4%
#64	Muse Glimmer 30B	Meta AI	80.2%
#65	Gemini 3.1 Flash Lite	Google DeepMind	80.1%
#66	Qwen3 235B A22B Thinking 2507	Alibaba Cloud	80.0%
#67	Gemini 2.5 Pro	Google DeepMind	79.9%
#68	GPT-5.2 Chat	OpenAI	79.8%
#69	GPT-5 Mini	OpenAI	79.6%
#70	Qwen3.6 27B	Alibaba Cloud	79.6%
#71	DeepSeek V3.1	DeepSeek	79.3%
#72	R1 0528	DeepSeek	79.1%
#73	GLM 4.6	Z.ai	78.8%
#74	DeepSeek V3.2 Exp	DeepSeek	78.8%
#75	GPT-5.4 Nano	OpenAI	77.9%
#76	Qwen3.5-9B	Alibaba Cloud	77.9%
#77	GLM 5	Z.ai	77.5%
#78	Claude Sonnet 4	Anthropic	76.7%
#79	Step 3.7 Flash	StepFun	76.6%
#80	MiMo-V2.5	Xiaomi	76.3%
#81	Kimi K2 0905	Moonshot AI	75.9%
#82	Mistral Small 4	Mistral AI	75.8%
#83	gpt-oss-120b	OpenAI	75.1%
#84	Auto Router (Beta)	OpenRouter	74.7%
#85	Qwen3 235B A22B Instruct 2507	Alibaba Cloud	74.4%
#86	Qwen3 Coder Next	Alibaba Cloud	74.1%
#87	Gemma 4 26B A4B	Google DeepMind	73.9%
#88	Ling 3.0 Flash	inclusionAI	73.0%
#89	Gemini 2.5 Flash	Google DeepMind	72.7%
#90	Claude Haiku 4.5	Anthropic	72.1%
#91	GPT-5 Nano	OpenAI	70.9%
#92	Qwen3 Next 80B A3B Instruct	Alibaba Cloud	70.7%
#93	Qwen3 VL 235B A22B Instruct	Alibaba Cloud	70.5%
#94	Nemotron 3.5 Lightning	NVIDIA	69.0%
#95	Gemma 4 26B A4B IT (free)	Google DeepMind	68.5%
#96	gpt-oss-20b	OpenAI	66.6%
#97	Llama 4 Maverick	Meta AI	65.9%
#98	GPT-4.1	OpenAI	64.9%
#99	Qwen3 VL 30B A3B Instruct	Alibaba Cloud	64.8%
#100	GLM 4.5 Air	Z.ai	64.5%
#101	GPT-4.1 Mini	OpenAI	64.1%
#102	Qwen3 30B A3B Instruct 2507	Alibaba Cloud	63.9%
#103	Qwen3 Coder 480B A35B	Alibaba Cloud	61.7%
#104	Qwen3 30B A3B	Alibaba Cloud	61.4%
#105	Qwen3 32B	Alibaba Cloud	61.4%
#106	DeepSeek V3	DeepSeek	60.9%
#107	Nemotron 3 Nano 30B A3B	NVIDIA	60.7%
#108	Qwen3 14B	Alibaba Cloud	59.4%
#109	DeepSeek V3 0324	DeepSeek	58.1%
#110	Gemini 2.5 Flash Lite	Google DeepMind	55.0%
#111	Claude Sonnet 4.6	Anthropic	54.4%
#112	Qwen3 Coder 30B A3B Instruct	Alibaba Cloud	52.0%
#113	Qwen3 VL 8B Instruct	Alibaba Cloud	51.8%
#114	GPT-4o (2024-08-06)	OpenAI	51.7%
#115	GLM 4.7 Flash	Z.ai	51.4%
#116	GPT-4.1 Nano	OpenAI	50.9%
#117	GPT-4o (2024-05-13)	OpenAI	50.7%
#118	GPT-4o	OpenAI	50.3%
#119	Llama 3.3 70B Instruct	Meta AI	49.5%
#120	Qwen 2.5 72B Instruct	Alibaba Cloud	44.6%
#121	Qwen2.5 VL 72B Instruct	Alibaba Cloud	44.4%
#122	GPT-4o mini	OpenAI	43.4%
#123	Qwen2.5 7B Instruct	Alibaba Cloud	32.8%
#124	Mistral Nemo	Mistral AI	31.6%
#125	Llama 3.1 8B Instruct	Meta AI	28.6%
#126	Llama 3 8B Lunaris	Sao10K	26.9%
#127	Ling 3.0 Flash Fin	inclusionAI	19.7%
#128	Llama 3.2 3B Instruct	Meta AI	11.6%
#129	Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)	Google DeepMind	10.3%

Source summary: “Gemini 3.1 Pro Preview ranks #1 and GPT-6 Astra #2, both at 94.4%, with Gemini 3.7 Flash third at 94.3% on OpenRouter's default-routing board.”

Publisher page renders a 129-row default-routing leaderboard; all 129 source-rendered rows are copied here in the publisher's own order. Provider-pinned sub-rows are not copied. Checked 2026-09-06.

Artificial Analysis Intelligence Index (v4.2 cohort)

Source: Artificial Analysis

v4.2

Publisher source

A composite benchmark aggregating ten challenging evaluations to provide a holistic measure of AI capabilities across mathematics, science, coding, and reasoning.

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#1	Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)	Anthropic	57
#2	GPT-6 Astra (max)	OpenAI	55
#3	GPT-6 Astra (xhigh)	OpenAI	54

Source summary: “Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) scores the highest on Artificial Analysis Intelligence Index with a score of 57, followed by GPT-6 Astra (max) with a score of 55, and GPT-6 Astra (xhigh) with a score of 54”

Publisher page reports 27 displayed results of 630 models and includes an ‘Estimate (independent evaluation forthcoming)’ label; only its source-rendered top-three summary is copied here. Historical v4.2 cohort: superseded by the publisher's v4.3 edition (2026-09-07) and kept unmodified for edition history. Checked 2026-09-05.

Artificial Analysis Intelligence Index (v4.3 cohort)

Source: Artificial Analysis

v4.3

Publisher source

Current-edition capture of the composite index (aggregating ten evaluations across agents, coding, general capability and scientific reasoning) under the publisher's v4.3 methodology, which changed agentic evaluation components. Scores are NOT comparable with the v4.2 cohort above: a methodology change is not a model change.

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#1	Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)	Anthropic	53
#2	Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Anthropic	53
#3	GPT-6 Astra (max)	OpenAI	53

Source summary: “Under v4.3, the publisher's rendered top three are tied at index 53: Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback), Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback) and GPT-6 Astra (max), in the publisher's rendered order.”

Publisher page reports 28 displayed results of 633 models; only its source-rendered top-three summary is copied here. The rendered order lists the tied rows 1–3; the tie is the publisher's rendering, not our resolution. Checked 2026-09-08.

Terminal-Bench v2.1

Source: Artificial Analysis

Terminus 2 · pass@1 · 3 repeats

Publisher source

A verified refresh of Terminal-Bench v2.0 — 89 curated tasks across software engineering, system administration, data processing, model training, and security, with environment and instruction fixes so scores reflect agent capability rather than environment gaps.

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#1	Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)	Anthropic	91.4%
#2	Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)	Anthropic	91.0%
#3	GPT-6 Astra (high)	OpenAI	89.9%

Source summary: “Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) scores the highest on Terminal-Bench v2.1 with a score of 91.4%, followed by Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback) with a score of 91.0%, and GPT-6 Astra (high) with a score of 89.9%”

Publisher page reports 27 displayed results of 235 models; only its source-rendered top-three summary is copied here. Checked 2026-09-05.

MMLU-Pro

Source: Artificial Analysis

Artificial Analysis current board

Publisher source

An enhanced version of MMLU with 12,000 graduate-level questions across 14 subject areas, featuring ten answer options and deeper reasoning requirements.

Publisher-rendered ranking rows
Rank	Model	Provider	Score
#1	Gemini 3 Pro Preview (high)	Google	89.8%
=2	Gemini 3 Pro Preview (low)	Google	89.5%
=2	Claude Opus 4.5 (Reasoning)	Anthropic	89.5%

Source summary: “Gemini 3 Pro Preview (high) scores the highest on MMLU-Pro with a score of 89.8%, followed by Gemini 3 Pro Preview (low) with a score of 89.5%, and Claude Opus 4.5 (Reasoning) with a score of 89.5%”

Publisher page reports 6 displayed results of 349 models; only its source-rendered top-three summary is copied here. Checked 2026-09-05.

Boards linked without copied scores

These sources did not expose ranking rows in server-rendered content, so this site does not guess or preserve an unchecked local copy.

LiveBench
The public page requires client-side JavaScript and did not expose a source-rendered ranking table to this snapshot process.
SWE-bench Verified
The source-rendered page documents the benchmark and harness variants but does not expose its interactive ranking rows.

Read benchmark methodology and limitations