whatmodel.sh / Claude Sonnet 4.6

Claude Sonnet 4.6

+ contribute

Claude Sonnet 4.6 is a closed-weights language model from Anthropic, released February 2026.

72.3±0.6#62 of 719 LLMsfrom 41 benchmark results across 22 sources

$ how it compares

The LLMs ranked either side of Claude Sonnet 4.6, out of 828 with a score, then more from Anthropic.

#	model	score	where	RAM	released
65	GPT-5 OpenAI	72±1	api $3.44/M		2025-08
66	Qwen3.6 Max Preview Alibaba	72±1	api $2.92/M		2026-04
67	Claude Sonnet 4.6 Anthropic	72±1	api $6.00/M		2026-02
68	Gemini 3 Flash (Preview) Google	72±1	api $1.13/M		2025-12
69	DeepSeek V4 Pro (Preview, Apr 2026) DeepSeek	72±1	local	~604GB	2026-04
	more from Anthropic
2	Claude Opus 5.5 Anthropic	87±1	api $8.00/M		2026-09
4	Claude Fable 5.1 Anthropic	85±1	api $20.00/M		2026-09
5	Claude Sonnet 5.5 Anthropic	84±1	api $4.00/M		2026-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #44 of 300 · 39% of its score · 12 tests

Beats Gemini 3.1 Pro (Preview), Muse Spark 1.2, Muse Spark 1.1 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro (split unspecified) lab	58.1% ✎	▄█·▄·▄██···▄ beats 0/9	very hard · avg 7.6%	▇▇▇▇ 10%	sharp · at its level
SWE-bench Verified	75.2% ✎	▁··▁·▁▂▁▄█▄▁ beats 18/29	hard · avg 19%	▇▇▇ 7.2%	sharp · at its level
Terminal-Bench 2.1 lab	67.0% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 31/56	hard · avg 17%	▇ 3.7%	at its level
OSWorld-Verified lab	78.5% ✎	█▄·▂▆▂▄▂▆▄▆█ beats 17/25	hard · avg 29%	▇ 3.2%	at its level
Agents' Last Exam overall full-pass rate (fixed OpenClaw + GUI-as-Tool)	9.9% ✎	▄···█▄▄▄█▄·█ beats 4/10	very hard · avg 1.7%	▇ 3.0%	typical
Humanity's Last Exam with tools lab	46.8% ✎	▁······▃·▄·█ beats 1/10	very hard · avg 14%	▇ 2.4%	at its level
Toolathlon (Tool Decathlon) lab	49.4% ✎	█···█··█·███ beats 0/5	hard · avg 16%	▇ 2.2%	at its level
Vibe Code Bench (test split, 50 web apps)	51.5% ✎	▃··▃·▃██▃▃▃█ beats 11/14	very hard · avg 2.3%	▇ 2.1%	sharp · at its level · noisy

4 more tests

test	result	vs the others in its run	difficulty	counts	why
Thinkingbox-bench pass@1 (507 stateful business-workflow agent tasks, N=20)	59.2% ✎	▅··▃·▃·█▃▅██ beats 12/15	very hard · avg 4.9%	▇ 1.7%	at its level
τ²-bench Telecom lab	97.9% ✎	▁·▅█▆▅▂▃▃▂▃▁ beats 80/89	hard · avg 29%	▇ 1.4%	sharp · noisy
Legal Research Bench (agentic, all-pass)	38.5% ✎	▄█··▄▄▄██·▄█ beats 10/12	very hard · avg 5.8%	▇ 1.3%	at its level
KlinikeBench strict pass@1 (333 interactive clinical encounters, Terminus-2)	8.4% ✎	▁▁·▁▁▁▁▅▃█▅█ beats 9/27	very hard · avg 2.4%	▇ 0.9%	typical

coding▁▁▁▃▆██▇▆▄▃▁ #45 of 458 · 28% of its score · 5 tests

Beats GPT-5.5, GPT-5.4, DeepSeek V4 Flash Vision (Exp) here, though they rank higher overall. Trails Mistral Large 4 (public preview) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro (split unspecified) lab	58.1% ✎	▄█·▄·▄██···▄ beats 0/9	very hard · avg 7.6%	▇▇▇▇ 10%	sharp · at its level
SWE-bench Verified	75.2% ✎	▁··▁·▁▂▁▄█▄▁ beats 18/29	hard · avg 19%	▇▇▇ 7.2%	sharp · at its level
Arena WebDev (Code Arena)	1521.4 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 78/119	avg 981	▇▇▇ 6.7%	sharp
Vibe Code Bench (test split, 50 web apps)	51.5% ✎	▃··▃·▃██▃▃▃█ beats 11/14	very hard · avg 2.3%	▇ 2.1%	sharp · at its level · noisy
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	74.5% ✎	▄··█·▄··▄█·▄ beats 7/7	hard · avg 34%	▇ 1.8%	at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #30 of 393 * · 3.8% of its score · 1 test

Beats Claude Sonnet 5.5, GPT-6 Sol, GPT-5.6 Terra here, though they rank higher overall. Trails Claude Opus 4.5 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1472.2 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 344/379	avg 1268	▇▇ 3.8%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #148 of 398 · 3.5% of its score · 3 tests

Trails Nemotron 3 Nano 30B-A3B, Qwen3-VL 8B Thinking, Olmo 3 32B Think here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
USAMO 2026 lab	55.0% ✎	█·██······██ beats 0/4	hard · avg 16%	▇ 1.4%	sharp · at its level · noisy
OTIS mock AIME 2024–2025	71.1% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 103/193	very hard · avg 8.5%	▇ 1.2%	sharp · noisy
MathConstraint full release, no tools (329 instances)	18.5% ✎	▄··▄▄·▄▄█▄·▄ beats 1/8	very hard · avg 11%	▇ 0.9%	at its level

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #31 of 147 · 11% of its score · 7 tests

Beats Claude Sonnet 5.5, MiMo-V2.6-Pro, MiMo-V2.6-Flash (309B-A15B) here, though they rank higher overall. Trails Qwen3.6 Plus, Qwen3.7 Plus, Gemini 3 Flash (Preview) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
CharXiv (reasoning questions) lab	71.6% ✎	███··█·██··█ beats 2/6	hard · avg 29%	▇ 2.7%	at its level
Arena vision (style controlled)	1275.2 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 119/145	avg 1068	▇ 2.6%	blunt
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	74.5% ✎	▄··█·▄··▄█·▄ beats 7/7	hard · avg 34%	▇ 1.8%	at its level
BEAR-Bench (bilingual RU/EN enterprise and academic document images, overall accuracy)	62.7% ✎	▃▃·▅·▃·█·█▅▅ beats 10/14	hard · avg 24%	▇ 1.3%	at its level
SciDocBench scientific-document workflows (overall, 496 instances)	49.6% ✎	▃▃···█▃▅▃·▃▅ beats 2/11	hard · avg 26%	▇ 1.3%	blunt · at its level
Arena document (style controlled)	1477.7 ✎	▁▁▃▃▄▃▇█▅▇▄▅ beats 28/37	avg 1335	▇ 1.3%	blunt
SpaceCast-Bench predictive spatial reasoning (overall, 16 task types)	36.0% ✎	▃·▅▃▅▃··▅··█ beats 3/11	very hard · avg 32%	▇ 0.4%	blunt · at its level · noisy

web search▂▁▂▆▄█▃▃▁▁·▁ #31 of 80 · 5.2% of its score · 3 tests

Beats Gemini 3.1 Pro (Preview), GPT-5.3-Codex, Qwen3.7 Max here, though they rank higher overall. Trails GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503), GPT-4o (2024-08-06) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam with tools lab	46.8% ✎	▁······▃·▄·█ beats 1/10	very hard · avg 14%	▇ 2.4%	at its level
Arena search systems (style controlled)	1201.0 ✎	▁·▁▁▂▂▂█▅▂▁▁ beats 21/30	avg 1055	▇ 1.5%	blunt
Legal Research Bench (agentic, all-pass)	38.5% ✎	▄█··▄▄▄██·▄█ beats 10/12	very hard · avg 5.8%	▇ 1.3%	at its level

other tests17 not in a task · 34% of its score

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam lab	34.6% ✎	▂·····▃·▄▂█▄ beats 1/10	very hard · avg 5.8%	▇▇ 4.3%	at its level
GPQA Diamond	78.8% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 123/208	hard · avg 40%	▇▇ 3.8%	sharp
Mystery Game Puzzles (Epoch)	14.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 20/74	very hard · avg 9.4%	▇ 3.4%	sharp · noisy
KINA (Knowledge Index of Noah's Ark) accuracy, 899 ten-option items	30.0% ✎	▁▁▂▁▃▄█▆▄▅▂▃ beats 20/39	very hard · avg 22%	▇ 2.8%	blunt · at its level
DSEffi-Bench Pass (1,000 data-science tasks, greedy)	59.7% ✎	▅··▃▅·█·▃██▃ beats 12/15	hard · avg 24%	▇ 2.6%	at its level
Chess Puzzles (Epoch)	3.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 17/135	very hard · avg 5.9%	▇ 2.5%	sharp · noisy
EQ-Bench 3 (emotional intelligence, judge Elo)	1384.2 ✎	▁·▂▁▄▅▅▃█▄▃▁ beats 61/62	avg 1044	▇ 2.3%	typical
The Token Games Elo (10-model puzzle-duel tournament)	1043.0 ✎	█··▄█··▄·▄·▄ beats 2/7	avg 815	▇ 2.2%	typical

9 more tests

test	result	vs the others in its run	difficulty	counts	why
AgentWorldBench (open-ended, overall)	56.0% ✎	▄▄··▄▄·▄█▄██ beats 7/11	hard · avg 25%	▇ 2.1%	blunt · at its level
Legal Research Bench (single-shot, no tools, all-pass)	7.3% ✎	▃▃█▃·▅···█▃▃ beats 11/12	very hard · avg 0.8%	▇ 1.5%	too hard for it
SimpleQA Verified	32.8% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 23/78	very hard · avg 8.0%	▇ 1.5%	at its level
LongSocialBench (full-context, 4-option MCQ over discussion threads)	44.3% ✎	▆▂·▄█·▃▂·▄▄▂ beats 10/16	very hard · avg 36%	▇ 1.4%	blunt · at its level
SkillsBench 87-task subset	54.4% ✎	▄··▄▄▄···█·█ beats 7/7	very hard · avg 8.2%	▇ 1.4%	at its level
WideSearch corrected dataset	79.5% ✎	▃······▃▃▃█▃ beats 7/7	hard · avg 20%	▇ 1.2%	at its level · noisy
DiagnosticIQ industrial maintenance MCQ (macro accuracy, 4 options)	64.8% ✎	▁··▃▁▁▁▅█▅▁▅ beats 16/24	medium · avg 52%	▇ 0.7%	blunt · at its level
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	89.4% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 38/98	very easy · avg 90%	▇ 0.0%	blunt
BrowseComp single-agent with context management	74.0% ✎	█··········█ beats 1/1		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · $6.00/M · $3 in / $15 out per 1M tokens · 1M context · docs.anthropic.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

$ how the score was worked out · Anthropic system cards and mo… 31%, Epoch AI — internal evaluatio… 20%, LMArena leaderboard dataset 16% and 19 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Anthropic system cards and model announcements (lab-reported scores) · lab-reported	9	▇▇▇▇▇▇▇▇▇ 31%	Factual vendor-reported score (not licensed content)
Epoch AI — internal evaluations	6	▇▇▇▇▇▇ 20%	CC BY 4.0
LMArena leaderboard dataset	5	▇▇▇▇▇ 16%	CC BY 4.0
Agents' Last Exam — authors' own fixed-harness model comparison	1	▇ 3%	CC BY 4.0
Knowledge Index of Noah's Ark (arXiv:2606.05104v2)	1	▇ 3%	CC BY 4.0
Legal Research Bench (Vals AI, arXiv:2610.00609v1)	2	▇ 3%	CC-BY-4.0
DSEffi-Bench: Demystifying Large Language Models' Capability in Efficient Data Science Code Generation — authors' own model runs	1	▇ 3%	CC BY 4.0
inclusionAI/Ling-3.0-flash model card	2	▇ 3%	MIT
EQ-Bench 3 canonical leaderboard results	1	▇ 2%	MIT
The Token Games: Evaluating Language Model Reasoning with Puzzle Duels — authors' own tournament	1	▇ 2%	CC BY-SA 4.0
Qwen/Qwen-AgentWorld-35B-A3B model card	1	▇ 2%	Apache-2.0
Vibe Code Bench (Vals AI, arXiv:2603.04601v3)	1	▇ 2%	CC-BY-4.0
VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)	1	▇ 2%	CC BY 4.0
One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows (arXiv:2608.19741v4)	1	▇ 2%	CC BY 4.0
LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads? (arXiv:2610.04118v1)	1	▇ 1%	CC BY 4.0
BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models (arXiv:2608.17895v1)	1	▇ 1%	CC BY 4.0
SciDocBench: A Workflow-Centered Benchmark and Data Pipeline for Scientific Document Understanding (arXiv:2609.05141v2)	1	▇ 1%	CC BY 4.0
MathConstraint: Automated Generation of Verified Combinatorial Reasoning Instances for LLMs — authors' own model runs	1	▇ 1%	CC BY 4.0
KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy (arXiv:2609.38480v2)	1	▇ 1%	CC BY 4.0
DiagnosticIQ: A Benchmark for LLM-Based Industrial Maintenance Action Recommendation from Symbolic Rules (arXiv:2605.08614v1)	1	▇ 1%	CC BY 4.0
SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models (arXiv:2610.12402v1)	1	▇ 0%	CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 72.3