Which AI model is best?

These rankings come from LMArena: users ask two anonymous models the same question and pick the better answer, and scores are calculated from millions of votes. They reflect real users' preferences, not developers' own claims.

Higher is better. The ± figure is the margin of error; models whose ranges overlap are effectively tied. "Prelim." marks models with few votes so far.

Overall (text)LMArena ↗

#	Model	Developer	Score	Votes
1	gemini-4-argon-highPrelim.	Google	1525 ±9	4,932
2	claude-opus-4-6-high	Anthropic	1505 ±3	77,636
3	claude-fable-5-high	Anthropic	1504 ±4	38,387
4	claude-opus-5.5-high	Anthropic	1504 ±9	4,552
5	claude-opus-4-7-high	Anthropic	1501 ±4	64,946
6	claude-fable-5.1-max	Anthropic	1501 ±6	11,800
7	claude-opus-4-6	Anthropic	1497 ±3	82,189
8	gemini-3.8-flash-highPrelim.	Google	1495 ±5	26,298
9	muse-spark-1.3-max	Meta	1494 ±6	12,343
10	claude-opus-4-7	Anthropic	1494 ±4	66,058

As of Oct 2, 2026 · 8,626,731 votes · 413 models

Open-weight models (text)LMArena ↗

#	Model	Developer	Score	Votes
1	kimi-k3-maxOpen	Moonshot	1488 ±5	28,280
2	mimo-v2.6-proOpen	Xiaomi	1480 ±9	4,056
3	glm-5.3-maxOpen	Z.ai	1478 ±6	17,857
4	glm-5.2-maxOpen	Z.ai	1476 ±4	45,399
5	deepseek-v4.1-flash-maxOpen	DeepSeek	1474 ±7	8,728
6	glm-5.3-flashOpen	Z.ai	1473 ±5	22,971
7	mimo-v2.5-proOpen	Xiaomi	1468 ±4	70,871
8	glm-5.1Open	Z.ai	1465 ±4	58,358
9	deepseek-v4-pro-high-20260813Open	DeepSeek	1464 ±7	9,780
10	kimi-k2.6Open	Moonshot	1461 ±4	39,936

As of Oct 2, 2026 · 8,626,731 votes · 413 models

CodingLMArena ↗

#	Model	Developer	Score	Votes
1	gemini-4-argon-highPrelim.	Google	1560 ±17	1,230
2	claude-fable-5-high	Anthropic	1552 ±7	10,038
3	claude-opus-4-6-high	Anthropic	1551 ±6	20,235
4	claude-opus-4-7-high	Anthropic	1551 ±6	18,521
5	claude-opus-4-6	Anthropic	1548 ±5	22,901
6	claude-opus-4-7	Anthropic	1546 ±6	18,728
7	gpt-6-astra-max	OpenAI	1543 ±13	2,135
8	gpt-6.1-sol-max	OpenAI	1542 ±24	609
9	kimi-k3-maxOpen	Moonshot	1541 ±8	7,285
10	mimo-v2.6-proOpen	Xiaomi	1540 ±18	1,103

As of Oct 2, 2026 · 1,848,296 votes · 408 models

Building web appsLMArena ↗

#	Model	Developer	Score	Votes
1	claude-opus-5.5-max	Anthropic	1815 +16/-16	2,062
2	gpt-6-astra-max	OpenAI	1788 +10/-10	6,123
3	claude-sonnet-5.5-xhigh	Anthropic	1786 +18/-18	1,531
4	gpt-6.1-sol-max	OpenAI	1758 +17/-17	1,620
5	claude-fable-5.1-max	Anthropic	1749 +10/-10	6,318
6	claude-sonnet-5.5-high	Anthropic	1715 +13/-13	2,527
7	claude-opus-5-max	Anthropic	1695 +6/-6	16,954
8	gpt-6-sol-max	OpenAI	1689 +11/-11	3,411
9	gemini-4-argon-highPrelim.	Google	1680 +13/-13	2,422
10	qwen3.8-maxPrelim.	Alibaba	1671 +12/-12	3,454

As of Oct 1, 2026 · 838,388 votes · 138 models

Understanding imagesLMArena ↗

#	Model	Developer	Score	Votes
1	claude-fable-5-high	Anthropic	1309 ±7	13,709
2	qwen3.8-max	Alibaba	1301 ±7	10,040
3	claude-opus-4-6-high	Anthropic	1299 ±6	22,328
4	claude-opus-4-7	Anthropic	1298 ±7	23,169
5	claude-opus-4-7-high	Anthropic	1298 ±7	22,755
6	gemini-3.7-flash-highPrelim.	Google	1296 ±10	3,814
7	claude-opus-4-6	Anthropic	1294 ±6	26,869
8	muse-spark	Meta	1294 ±9	5,868
9	muse-spark-1.2 (xHigh)	Meta	1293 ±14	1,951
10	gpt-6.1-sol-max	OpenAI	1291 ±16	1,408

As of Oct 2, 2026 · 1,436,923 votes · 159 models

Web searchLMArena ↗

#	Model	Developer	Score	Votes
1	gpt-5.6-sol-xhigh	OpenAI	1257 ±7	29,663
2	claude-opus-4-6-search	Anthropic	1253 ±5	134,699
3	gpt-5.5-search	OpenAI	1242 ±5	89,873
4	claude-opus-4-7	Anthropic	1233 ±5	91,394
5	claude-fable-5-high	Anthropic	1230 ±8	41,795
6	ernie-5.1	Baidu	1227 ±10	3,788
7	claude-sonnet-4-6-search	Anthropic	1221 ±5	134,905
8	grok-4.5	SpaceXAI	1213 ±7	31,505
9	gemini-3.1-pro-grounding	Google	1210 ±5	113,282
10	gemini-3-pro-grounding	Google	1207 ±6	37,024

As of Aug 24, 2026 · 1,110,523 votes · 34 models

Image generationLMArena ↗

#	Model	Developer	Score	Votes
1	gpt-image-2.5-sunburstPrelim.	OpenAI	1425 ±7	16,652
2	gpt-image-2.5-flarePrelim.	OpenAI	1397 ±7	15,792
3	gpt-image-2 (medium)	OpenAI	1384 ±4	92,549
4	grok-imagine-image-2.0 (canvas)Prelim.	SpaceXAI	1336 ±12	3,070
5	mai-image-2.6	Microsoft AI	1333 ±5	23,918
6	reve-2.1	Reve	1302 ±8	8,250
7	grok-imagine-image-2.0 (20260801)	SpaceXAI	1297 ±6	13,278
8	muse-image	Meta	1274 ±5	41,071
9	reve-2.0	Reve	1269 ±6	15,774
10	gemini-3.1-flash-image (nano-banana-2) [web-search]	Google	1261 ±4	59,301

As of Oct 5, 2026 · 6,583,920 votes · 81 models

Image editingLMArena ↗

#	Model	Developer	Score	Votes
1	gpt-image-2.5-sunburstPrelim.	OpenAI	1524 ±5	58,172
2	gpt-image-2.5-flarePrelim.	OpenAI	1481 ±5	55,745
3	gpt-image-2 (medium)	OpenAI	1462 ±3	304,260
4	grok-imagine-image-2.0 (canvas)Prelim.	SpaceXAI	1437 ±9	10,821
5	mai-image-2.6	Microsoft AI	1427 ±5	50,147
6	grok-imagine-image-2.0 (20260801)	SpaceXAI	1426 ±5	48,559
7	muse-image	Meta	1403 ±4	141,017
8	mai-image-2.5	Microsoft AI	1401 ±4	190,078
9	seedream-5.0-pro	Bytedance	1394 ±3	281,170
10	grok-imagine-image-quality (20260519)	SpaceXAI	1391 ±6	38,195

As of Oct 5, 2026 · 30,962,085 votes · 58 models

Video generationLMArena ↗

#	Model	Developer	Score	Votes
1	gemini-omni-1.1-flash	Google	1516 ±15	1,784
2	gemini-omni-flash	Google	1513 ±9	26,576
3	flux-3-videoPrelim.	Black Forest Labs	1493 ±17	1,302
4	grok-imagine-video-1.5-agentPrelim.	SpaceXAI	1492 ±18	1,218
5	dreamina-seedance-2.0-720p	Bytedance	1479 ±8	56,318
6	wan3.0	Alibaba	1476 ±13	2,598
7	dreamina-seedance-2.5-720p	Bytedance	1474 ±9	8,001
8	minimax-h3Open	MiniMax	1460 ±9	11,417
9	muse-video	Meta	1456 ±15	2,188
10	happyhorse-1.0	Alibaba-ATH	1427 ±13	22,264

As of Sep 21, 2026 · 718,577 votes · 48 models

Model releases

Figures from developers' official announcements or the cited reports. Scores are as published by the developer at launch. Test conditions differ between companies, so compare with care.

Date	Model	Developer	API price	Reported results (selected)
7 Oct 2026	Claude Haiku 5.5 cuts prices 90% to $0.10 per million tokens and beats the previous Sonnet on office work	Anthropic	$0.10 / $0.50 per million input / output tokens (prompts up to 100k; $0.50 / $2.50 above)	—
30 Sept 2026	Google announces Gemini 4 Argon	Google DeepMind	Introductory $2 / $10 per million input / output tokens, later $4 / $20	DeepSWE v1.1: 77.9% AutomationBench: 51.3%
28 Sept 2026	Claude Sonnet 5.5 nears Opus 5.5 at half the price, scoring 70.6% on a coding test where Sonnet 5 managed 10.3%	Anthropic	$2 / $10 per million input / output tokens	—
22 Sept 2026	Claude Opus 5.5	Anthropic	$4 / $20 per million input / output tokens; cache reads $0.20	Terminal-Bench 4.0: 66.4% Humanity's Last Exam (with tools): 67.7%
3 Sept 2026	OpenAI releases GPT-6 Astra	OpenAI	—	ARC-AGI-3: 99.9%
1 Sept 2026	Claude Fable 5.1 and Mythos 5.1	Anthropic	$10 / $50 per million input / output tokens; cache reads $0.25 (75% lower)	—
24 Jul 2026	Claude Opus 5	Anthropic	$5 / $25 per million input / output tokens	—
9 Jun 2026	Claude Fable 5 and Mythos 5	Anthropic	$10 / $50 per million input / output tokens	—
24 Apr 2026	DeepSeek V4: open weights with a 1-million-token context	DeepSeek	—	V4-Pro: 1.6T total / 49B active parameters. V4-Flash: 284B total / 13B active. 1M-token context.
5 Feb 2026	Claude Opus 4.6 with a 1-million-token context	Anthropic	$5 / $25 per million input / output tokens	1M-token context window (beta)
24 Nov 2025	Claude Opus 4.5, at a lower price	Anthropic	$5 / $25 per million input / output tokens	—
18 Nov 2025	Google releases Gemini 3	Google DeepMind	—	Humanity's Last Exam (no tools): 37.5% GPQA Diamond: 91.9%
29 Sept 2025	Claude Sonnet 4.5	Anthropic	—	SWE-bench Verified: 77.2% OSWorld: 61.4%
7 Aug 2025	OpenAI releases GPT-5	OpenAI	—	SWE-bench Verified: 74.9% AIME 2025 (no tools): 94.6%
22 May 2025	Anthropic releases Claude 4	Anthropic	—	SWE-bench Verified: 72.5% Terminal-bench: 43.2%
5 Apr 2025	Meta releases Llama 4	Meta	—	Scout: 17B active parameters, 16 experts, 10M-token context. Maverick: 17B active, 128 experts.
25 Mar 2025	Google releases Gemini 2.5, its first "thinking" model	Google DeepMind	—	Humanity's Last Exam (no tools): 18.8%
24 Feb 2025	Claude 3.7 Sonnet and Claude Code	Anthropic	$3 / $15 per million input / output tokens	SWE-bench Verified: 63.7%
20 Jan 2025	DeepSeek releases R1, an open reasoning model	DeepSeek	—	AIME 2024 (pass@1): 79.8% MATH-500 (pass@1): 97.3%