Leaderboard

What the AI Models Cost in 2026

78 models compared on price per million tokens, context window and release date. Public arena ELO is shown for the 70 models that have one; the rest are listed by release date, because a model without enough arena votes has no honest rank to give. Newest here: September 2026.

Search models

78 models — ranked by ELO, newest first

Ranked list shows models with public ELO data. New models without enough Arena votes are listed separately below.

#ModelDeveloperELOMMLUContextPrice (Input)Model IDOfficialType
1Claude Fable 5Anthropic1506-1M$10.00claude-fable-5Claude models docs ->Closed
2Claude Opus 4.6Anthropic150591.11M$5.00claude-opus-4.6Claude models docs ->Closed
3Claude Opus 4.7Anthropic1502-1M$5.00claude-opus-4.7Claude models docs ->Closed
4Claude Fable 5.1Anthropic1498-1M$10.00claude-fable-5-1Claude models docs ->Closed
5Claude Opus 5Anthropic1493-1M$5.00claude-opus-5Claude models docs ->Closed
6Gemini 3.8 FlashGoogle1493-1M$0.75gemini-3-8-flashGemini models docs ->Closed
7Muse Spark 1.3Meta1493-1M$1.25muse-spark-1-3Llama model hub ->Closed
8Gemini 3.7 FlashGoogle1490-1M$0.75gemini-3-7-flashGemini models docs ->Closed
9Gemini 3.1 ProGoogle1487-1M$2.00gemini-3.1-proGemini models docs ->Closed
10Kimi K3Moonshot AI1485-1M$3.00kimi-k3Moonshot API docs ->Open
11Gemini 3 ProGoogle148591.81Mgemini-3-proGemini models docs ->Closed
12GPT-5.6 SolOpenAI1483-1.05M$4.00gpt-5-6-solOpenAI models docs ->Closed
13GLM 5.3Z.ai1483-1M$1.40glm-5-3Find official docs ->Closed
14GPT-5.5OpenAI1482-1M (API) / 400K (Codex)$5.00gpt-5.5OpenAI models docs ->Closed
15Claude Opus 4.8Anthropic1481-1M$5.00claude-opus-4.8Claude models docs ->Closed
16Qwen3.8 MaxAlibaba1481-1M$2.00qwen3-8-maxQwen docs ->Open
17Gemini 3.6 FlashGoogle1480-1M$0.75gemini-3-6-flashGemini models docs ->Closed
18GPT-6 AstraOpenAI1480-1.05M$10.00gpt-6-astraOpenAI models docs ->Closed
19GPT-5.4OpenAI1476-1M$2.50gpt-5.4OpenAI models docs ->Closed
20Grok 4.20xAI1475-256K$3.00grok-4-20xAI API docs ->Closed
21GLM 5.3 FlashZ.ai1475-1M$0.15glm-5-3-flashFind official docs ->Open
22GPT-5.5 InstantOpenAI1474-128K$2.50gpt-5-5-instantOpenAI models docs ->Closed
23Claude Opus 4.5Anthropic147390.8200K$5.00claude-opus-4.5Claude models docs ->Closed
24Claude Sonnet 4.6Anthropic147389.3200K$3.00claude-sonnet-4.6Claude models docs ->Closed
25GLM 5.2Z.ai1472-1M$0.49glm-5-2Find official docs ->Open
26GPT-5.6 TerraOpenAI1466-1.05M$2.00gpt-5-6-terraOpenAI models docs ->Closed
27GLM 5.1Z.ai1466-128K$1.00glm-5-1Find official docs ->Closed
28DeepSeek V4 ProDeepSeek1463-1M$1.32deepseek-v4-proDeepSeek API docs ->Closed
29Claude Sonnet 5Anthropic1461-1M$2.00claude-sonnet-5Claude models docs ->Closed
30Kimi K2.6Moonshot AI1460-256K$1.50kimi-k2-6Moonshot API docs ->Closed
31GLM 5Z.ai1458-128K$0.80glm-5Find official docs ->Closed
32Grok 4.6xAI1456-500K$2.00grok-4-6xAI API docs ->Closed
33GPT-5.6 LunaOpenAI1452-1.05M$0.20gpt-5-6-lunaOpenAI models docs ->Closed
34Kimi K2.5Moonshot AI1450-262Kkimi-k2.5Moonshot API docs ->Closed
35GPT-5.4 MiniOpenAI1448-400K$0.75gpt-5.4-miniOpenAI models docs ->Closed
36Gemini 2.5 ProGoogle1445-1M$1.25gemini-2.5-proGemini models docs ->Closed
37GPT-4.5 PreviewOpenAI1445-128K$75.00gpt-4.5-previewOpenAI models docs ->Closed
38Grok 4.3xAI1443-N/Agrok-4.3xAI API docs ->Closed
39DeepSeek V4 FlashDeepSeek1438-128Kdeepseek-v4-flashDeepSeek API docs ->Closed
40GPT-5.2OpenAI143889.6400K$1.75gpt-5.2OpenAI models docs ->Closed
41Qwen3.8 27BAlibaba1437-262K$0.50qwen3-8-27bQwen docs ->Open
42o3OpenAI1432-200K$10.00o3OpenAI models docs ->Closed
43Gemini 3.1 Flash-LiteGoogle1432-1M$0.25gemini-3-1-flash-liteGemini models docs ->Closed
44Claude Opus 4Anthropic1426-200K$5.00claude-opus-4Claude models docs ->Closed
45DeepSeek V3.2DeepSeek1425-128K$0.27deepseek-v3-2DeepSeek API docs ->Open
46DeepSeek R1DeepSeek142190.864K$0.55deepseek-r1DeepSeek API docs ->Open
47Claude Haiku 4.5Anthropic1415-200K$1.00claude-haiku-4.5Claude models docs ->Closed
48Mistral Large 3Mistral AI1413-128K$2.00mistral-large-3Mistral models docs ->Open
49Grok-3xAI141292.7131K$3.00grok-3xAI API docs ->Closed
50Gemini 2.5 FlashGoogle1410-1M$0.30gemini-2.5-flashGemini models docs ->Closed
51o1OpenAI140290.8200K$15.00o1OpenAI models docs ->Closed
52GPT-5.4 NanoOpenAI1402-400K$0.20gpt-5.4-nanoOpenAI models docs ->Closed
53Claude Sonnet 4Anthropic1401-200K$3.00claude-sonnet-4Claude models docs ->Closed
54DeepSeek V3DeepSeek139688.5128K$0.14deepseek-v3DeepSeek API docs ->Open
55Claude 3.5 Sonnet (Oct 2024)Anthropic137490.4200K$3.00claude-3.5-sonnet-octClaude models docs ->Closed
56o3-miniOpenAI1363-200K$1.10o3-miniOpenAI models docs ->Closed
57Gemini 2.0 FlashGoogle136076.41M$0.10gemini-2.0-flashGemini models docs ->Closed
58Gemini 1.5 ProGoogle135185.92M$1.25gemini-1.5-proGemini models docs ->Closed
59GPT-4oOpenAI134688.7128K$2.50gpt-4oOpenAI models docs ->Closed
60Grok-2xAI1336-128K$2.00grok-2xAI API docs ->Closed
61Llama 3.1 405BMeta133586128K$0.80llama-3.1-405bLlama model hub ->Open
62Llama 4 MaverickMeta1327-N/Allama-4-maverickLlama model hub ->Open
63Claude 3 OpusAnthropic132286.8200K$15.00claude-3-opusClaude models docs ->Closed
64Llama 4 ScoutMeta1321-N/Allama-4-scoutLlama model hub ->Open
65GPT-4o MiniOpenAI131882128K$0.15gpt-4o-miniOpenAI models docs ->Closed
66Mistral Large 2Mistral AI131484128K$2.00mistral-large-2Mistral models docs ->Closed
67Qwen 2.5 72B InstructAlibaba130385.3128K$0.30qwen-2.5-72bQwen docs ->Open
68Llama 3.1 70BMeta129383.6128K$0.35llama-3.1-70bLlama model hub ->Open
69Claude 3 HaikuAnthropic126175.2200K$0.25claude-3-haikuClaude models docs ->Closed
70Llama 3.1 8BMeta121173128K$0.05llama-3.1-8bLlama model hub ->Open

New Models (Awaiting Public ELO)

ModelDeveloperReleasedContextPrice (Input)Official
Claude Opus 5.5Anthropic2026-091M$4.00Claude models docs ->
GPT-6 SolOpenAI2026-091.05M$2.00OpenAI models docs ->
GPT-6 LunaOpenAI2026-091.05M$0.10OpenAI models docs ->
Claude Mythos 5.1Anthropic2026-091M$10.00Claude models docs ->
DeepSeek V4.1 FlashDeepSeek2026-091M$0.30DeepSeek API docs ->
Seed 2.1 TurboByteDance2026-08262K$0.50Find official docs ->
Qwen3.8 FlashAlibaba2026-08262K$0.16Qwen docs ->
GPT-5.5 ProOpenAI2026-05256K$5.00OpenAI models docs ->

Model Profiles

#1Proprietary

Claude Fable 5

Anthropic

Anthropic's first generally available Mythos-class model (June 9, 2026), restored globally after the June access pause. Built for long-horizon autonomous work, advanced coding, vision, memory, and professional tasks. Priced at $10/$50 per million tokens, with safety routing for high-risk requests. Arena text score 1506 (claude-fable-5-high, 30,057 votes), leaderboard of 13 September 2026.

1506

ELO

1M

Context

$10.00

per 1M tokens

#2Proprietary

Claude Opus 4.6

Anthropic

A long-time leader of the Arena text leaderboard, with 99.8% AIME 2025 and 80.8% SWE-bench, leading in coding and hard prompts. Arena text score 1505 (claude-opus-4-6-high, 71,993 votes), leaderboard of 13 September 2026.

1505

ELO

1M

Context

$5.00

per 1M tokens

#3Proprietary

Claude Opus 4.7

Anthropic

Anthropic's most capable generally available model, launched in April 2026 with stronger long-horizon agentic performance and the same $5/$25 MTok pricing as Opus 4.6. Arena text score 1502 (claude-opus-4-7-high, 60,002 votes), leaderboard of 13 September 2026.

1502

ELO

1M

Context

$5.00

per 1M tokens

#4Proprietary

Claude Fable 5.1

Anthropic

Anthropic's flagship, released 1 September 2026. Input and output prices are unchanged from Fable 5, but cache reads fall from $1.00 to $0.25 per million, a 75% cut that makes repeated long-context work far cheaper than the headline price suggests. Cache writes $12.50 for five minutes, $20.00 for an hour. 1M-token context, up to 128K tokens per answer. Prices checked 1 September 2026. Arena text score 1498 (claude-fable-5.1-max, 5,783 votes), leaderboard of 13 September 2026.

1498

ELO

1M

Context

$10.00

per 1M tokens

#5Proprietary

Claude Opus 5

Anthropic

Anthropic's July 2026 Opus release. Stronger than Opus 4.8 on coding, knowledge work, long-horizon agentic tasks, and cost per completed task. Same $5/$25 per million token base pricing, with Fast mode available. Arena text score 1493 (claude-opus-5-high, 42,617 votes), leaderboard of 13 September 2026.

1493

ELO

1M

Context

$5.00

per 1M tokens

#6Proprietary

Gemini 3.8 Flash

Google

Google's third Flash release in six weeks, shipped 2 September 2026 — a better model at the same price as 3.7 Flash, with a 1M-token context and 64K output. The $0.75 input / $3.75 output rate is introductory and doubles to $1.50 / $7.50 on 1 January 2027. Prices checked 3 September 2026. Arena text score 1493 (gemini-3.8-flash-high, 5,076 votes), leaderboard of 13 September 2026.

1493

ELO

1M

Context

$0.75

per 1M tokens

#7Proprietary

Muse Spark 1.3

Meta

Meta's reasoning model, released early September 2026 and delivered through Muse Code and the Meta Model API. A 1M-token context with text, image, and video input; it lands around #6 on the Artificial Analysis Intelligence Index. The standard endpoint is $1.25 input / $4.25 output per million (cached input $0.15); a contributor tier runs $0.10 / $0.20 if Meta may train on your prompts, and a higher 'max reasoning' tier is in limited preview. Prices checked 3 September 2026. Arena text score 1493 (muse-spark-1.3-max, 4,723 votes, still preliminary), leaderboard of 13 September 2026.

1493

ELO

1M

Context

$1.25

per 1M tokens

#8Proprietary

Gemini 3.7 Flash

Google

Google's August 2026 Flash model with a 1M-token context and 64K output. The listed price is introductory and runs to 31 December 2026, after which it doubles to $1.50 input and $7.50 output per million. Prices checked 1 September 2026. Arena text score 1490 (gemini-3.7-flash-high, 5,640 votes), leaderboard of 13 September 2026.

1490

ELO

1M

Context

$0.75

per 1M tokens

#9Proprietary

Gemini 3.1 Pro

Google

Google's newer Pro generation surfaced at Cloud Next 2026 as its most capable model for complex workflows, with 1M context and Gemini 3-class multimodal support. Arena text score 1487 (gemini-3.1-pro-preview, 106,951 votes), leaderboard of 13 September 2026.

1487

ELO

1M

Context

$2.00

per 1M tokens

#10Open Source

Kimi K3

Moonshot AI

The first open-weight model in the ~3T parameter class (July 16, 2026): a 2.8-trillion-parameter mixture-of-experts with native vision and a 1M-token context. Cached input drops to $0.30 per million, which makes repeated long-context work unusually cheap. Arena text score 1485 (kimi-k3-max, 20,987 votes), leaderboard of 13 September 2026.

1485

ELO

1M

Context

$3.00

per 1M tokens

#11Proprietary

Gemini 3 Pro

Google

Google's late-2025 Pro model with 94.3% GPQA and 100% AIME 2025, a 1M context window, and strong multimodal reasoning for complex workflows. Arena text score 1485 (gemini-3-pro, 40,654 votes), leaderboard of 13 September 2026.

1485

ELO

1M

Context

per 1M tokens

#12Proprietary

GPT-5.6 Sol

OpenAI

Flagship tier of the GPT-5.6 family (July 9, 2026). Scores 94.6% on GPQA Diamond, 90.4% on BrowseComp and 89% on FrontierMath Tier 1-3. Price cut on 21 August 2026 from $5/$30 to $4/$20, a promotional rate that runs through at least 21 November 2026. 922K input / 128K output; requests above 272K input tokens are billed at 2x input and 1.5x output. Arena text score 1483 (gpt-5.6-sol-xhigh, 27,069 votes), leaderboard of 13 September 2026.

1483

ELO

1.05M

Context

$4.00

per 1M tokens

#13Proprietary

GLM 5.3

Z.ai

Z.ai's coding and cyber-defence model, released 14 August 2026. It reuses the GLM 5.2 base unchanged and takes its reported gains from post-training alone. Open weights were promised about two weeks after launch and had not appeared as of 1 September 2026. Cached input $0.26 per million. Prices checked 1 September 2026. Arena text score 1483 (glm-5.3-max, 10,960 votes), leaderboard of 13 September 2026.

1483

ELO

1M

Context

$1.40

per 1M tokens

#14Proprietary

GPT-5.5

OpenAI

OpenAI's April 2026 flagship model for real-world coding and professional workflows, with stronger agentic performance and a 1M API context window. Arena text score 1482 (gpt-5.5-high, 64,924 votes), leaderboard of 13 September 2026.

1482

ELO

1M (API) / 400K (Codex)

Context

$5.00

per 1M tokens

#15Proprietary

Claude Opus 4.8

Anthropic

Anthropic's May 2026 Opus upgrade. 4x less likely to overlook code flaws than Opus 4.7. First model to complete every case on Super-Agent benchmark. Highest score ever on Legal Agent Benchmark at launch. Fast mode available at $10/$50 per million tokens. Arena text score 1481 (claude-opus-4-8-high, 52,535 votes), leaderboard of 13 September 2026.

1481

ELO

1M

Context

$5.00

per 1M tokens

#16Open Source

Qwen3.8 Max

Alibaba

Alibaba's 2.4-trillion-parameter mixture-of-experts, launched 3 August 2026 and released with open weights. Takes text, images and video, answers up to 128K tokens at a time, and cached input drops to $0.25 per million. Prices checked 1 September 2026. Arena text score 1481 (qwen3.8-max, 16,670 votes), leaderboard of 13 September 2026.

1481

ELO

1M

Context

$2.00

per 1M tokens

#17Proprietary

Gemini 3.6 Flash

Google

Google's speed-and-value play against GPT-5.6 (July 21, 2026). 1M-token context, knowledge cutoff March 2026, and the strongest price-per-task ratio in the current flagship group for high-throughput work. Priced at $0.75/$3.75 through 31 December 2026, rising to $1.50/$7.50 on 1 January 2027. Prices checked 21 September 2026. Arena text score 1480 (gemini-3.6-flash-high, 26,445 votes), leaderboard of 13 September 2026.

1480

ELO

1M

Context

$0.75

per 1M tokens

#18Proprietary

GPT-6 Astra

OpenAI

OpenAI's new frontier model, released 3 September 2026 to partners and rolled out publicly the next day. Scores 96% on GPQA Diamond and 74.1% on DeepSWE v1.1, with the real gains on computer use, terminal work, cybersecurity and long multi-step tasks rather than the saturated reasoning benchmarks. It introduces recurrent depth, a reasoning method that hides part of the chain of thought, which is why the release drew safety scrutiny. 1.05M-token context, up to 128K tokens per answer, knowledge cutoff April 2026. Long-context requests are billed at $20 input and $75 output per million. Prices checked 5 September 2026. Arena text score 1480 (gpt-6-astra-max, 2,693 votes, still preliminary), leaderboard of 13 September 2026.

1480

ELO

1.05M

Context

$10.00

per 1M tokens

#19Proprietary

GPT-5.4

OpenAI

OpenAI's March 2026 frontier model for professional reasoning and agentic coding, released across ChatGPT, API, and Codex. Arena text score 1476 (gpt-5.4-high, 60,537 votes), leaderboard of 13 September 2026.

1476

ELO

1M

Context

$2.50

per 1M tokens

#20Proprietary

Grok 4.20

xAI

xAI's latest model with improved reasoning and coding capabilities. Arena text score 1475 (grok-4.20-beta1, 26,599 votes), leaderboard of 13 September 2026.

1475

ELO

256K

Context

$3.00

per 1M tokens

#21Open Source

GLM 5.3 Flash

Z.ai

Open weights under MIT, released 26 August 2026, with a 1M-token context at fifteen cents per million input. A launch promotion halved input to $0.075 until 9 September 2026. Prices checked 1 September 2026. Arena text score 1475 (glm-5.3-flash, 10,038 votes), leaderboard of 13 September 2026.

1475

ELO

1M

Context

$0.15

per 1M tokens

#22Proprietary

GPT-5.5 Instant

OpenAI

Fast default model powering ChatGPT for all users. Optimized for speed while maintaining GPT-5.5 quality. Arena text score 1474 (gpt-5.5-instant, 25,850 votes), leaderboard of 13 September 2026.

1474

ELO

128K

Context

$2.50

per 1M tokens

#23Proprietary

Claude Opus 4.5

Anthropic

Major upgrade with 87% GPQA and 80.9% SWE-bench, the highest-rated Anthropic model before the 4.6 generation. Arena text score 1473 (claude-opus-4-5-20251101-high-32k, 36,239 votes), leaderboard of 13 September 2026.

1473

ELO

200K

Context

$5.00

per 1M tokens

#24Proprietary

Claude Sonnet 4.6

Anthropic

Latest Sonnet with adaptive reasoning, 89.9% GPQA and 79.6% SWE-bench, excellent balance of speed and intelligence. Arena text score 1473 (claude-sonnet-4-6, 66,208 votes), leaderboard of 13 September 2026.

1473

ELO

200K

Context

$3.00

per 1M tokens

#25Open Source

GLM 5.2

Z.ai

The open-weight Chinese model that made the case that a frontier-class base need not be expensive, released 16 June 2026 with a 1M-token context. Still the base underneath GLM 5.3. Prices checked 1 September 2026. Arena text score 1472 (glm-5.2-max, 36,798 votes), leaderboard of 13 September 2026.

1472

ELO

1M

Context

$0.49

per 1M tokens

#26Proprietary

GPT-5.6 Terra

OpenAI

Balanced tier of the GPT-5.6 family: most of Sol's capability at 40% of the price. OpenAI cut Terra pricing 20% on July 30, 2026. The default choice for production workloads that don't need frontier reasoning. Arena text score 1466 (gpt-5.6-terra-xhigh, 28,119 votes), leaderboard of 13 September 2026.

1466

ELO

1.05M

Context

$2.00

per 1M tokens

#27Proprietary

GLM 5.1

Z.ai

Chinese frontier model from Zhipu AI. Competitive with Claude Sonnet at a fraction of the cost. Arena text score 1466 (glm-5.1, 48,901 votes), leaderboard of 13 September 2026.

1466

ELO

128K

Context

$1.00

per 1M tokens

#28Proprietary

DeepSeek V4 Pro

DeepSeek

DeepSeek's higher-capability V4-generation API model introduced in April 2026 for deeper reasoning and agentic workloads. Listed prices are peak-hour rates (01:00-04:00 and 06:00-10:00 UTC on weekdays); off-peak costs half, $0.66/$1.98. 1M context, up to 384K output. Prices checked 21 September 2026. Arena text score 1463 (deepseek-v4-pro-high-20260813, 9,008 votes), leaderboard of 13 September 2026.

1463

ELO

1M

Context

$1.32

per 1M tokens

#29Proprietary

Claude Sonnet 5

Anthropic

Anthropic's most agentic Sonnet yet (June 30, 2026), approaching Opus 4.8 quality at a fraction of the cost. Best for agentic coding, multi-file refactors, long-document analysis and computer use. Launched at $2/$10 as introductory pricing; on 10 August 2026 Anthropic made that the permanent price and cancelled the planned rise to $3/$15. Prices checked 21 September 2026. Arena text score 1461 (claude-sonnet-5-high, 35,301 votes), leaderboard of 13 September 2026.

1461

ELO

1M

Context

$2.00

per 1M tokens

#30Proprietary

Kimi K2.6

Moonshot AI

Moonshot AI's latest model with strong multilingual and long-context capabilities. $20B valuation. Arena text score 1460 (kimi-k2.6, 37,502 votes), leaderboard of 13 September 2026.

1460

ELO

256K

Context

$1.50

per 1M tokens

#31Proprietary

GLM 5

Z.ai

Previous generation Zhipu model, still competitive with Western mid-tier models. Arena text score 1458 (glm-5, 27,605 votes), leaderboard of 13 September 2026.

1458

ELO

128K

Context

$0.80

per 1M tokens

#32Proprietary

Grok 4.6

xAI

xAI's August 2026 flagship, built for long-running agents and coding. Watch the billing cliff: once a prompt reaches 200K tokens the whole request is charged at $4.00 input and $12.00 output per million, not just the tokens above the line. Prices checked 1 September 2026. Arena text score 1456 (grok-4.6-high, 15,521 votes), leaderboard of 13 September 2026.

1456

ELO

500K

Context

$2.00

per 1M tokens

#33Proprietary

GPT-5.6 Luna

OpenAI

Fastest and cheapest GPT-5.6 tier, cut 80% in price on July 30, 2026. At $0.20 per million input tokens it is priced for high-volume classification, routing and extraction rather than deep reasoning. Arena text score 1452 (gpt-5.6-luna-xhigh, 28,547 votes), leaderboard of 13 September 2026.

1452

ELO

1.05M

Context

$0.20

per 1M tokens

#34Proprietary

Kimi K2.5

Moonshot AI

Chinese model with the highest HumanEval score ever recorded (99.0%), excelling at code generation and reasoning. Arena text score 1450 (kimi-k2.5-thinking, 70,513 votes), leaderboard of 13 September 2026.

1450

ELO

262K

Context

per 1M tokens

#35Proprietary

GPT-5.4 Mini

OpenAI

Faster and lower-cost GPT-5.4 variant for high-volume coding, subagents, and computer-use workloads. Arena text score 1448 (gpt-5.4-mini-high, 59,387 votes), leaderboard of 13 September 2026.

1448

ELO

400K

Context

$0.75

per 1M tokens

#36Proprietary

Gemini 2.5 Pro

Google

Google's hybrid thinking model combining fast responses with deep reasoning, top performer on coding and math benchmarks. Arena text score 1445 (gemini-2.5-pro, 122,554 votes), leaderboard of 13 September 2026.

1445

ELO

1M

Context

$1.25

per 1M tokens

#37Proprietary

GPT-4.5 Preview

OpenAI

OpenAI's largest and most knowledgeable non-reasoning model with broad world knowledge and reduced hallucinations. Arena text score 1445 (gpt-4.5-preview-2025-02-27, 14,547 votes), leaderboard of 13 September 2026.

1445

ELO

128K

Context

$75.00

per 1M tokens

#38Proprietary

Grok 4.3

xAI

xAI's newer flagship family which ranked near the top of LMArena at launch in both thinking and non-thinking modes. Arena text score 1443 (grok-4.3, 66,801 votes), leaderboard of 13 September 2026.

1443

ELO

N/A

Context

per 1M tokens

#39Proprietary

DeepSeek V4 Flash

DeepSeek

DeepSeek's April 2026 V4-generation API model focused on speed and lower-cost production inference. Retired in September 2026: API requests under this name are now served by DeepSeek V4.1 Flash at Flash pricing. Arena text score 1438 (deepseek-v4-flash-high-preview, 48,597 votes), leaderboard of 13 September 2026.

1438

ELO

128K

Context

per 1M tokens

#40Proprietary

GPT-5.2

OpenAI

OpenAI's current-gen flagship with 400K context, 92.4% GPQA and 100% AIME 2025, strong reasoning at reduced cost. Arena text score 1438 (gpt-5.2-high, 47,538 votes), leaderboard of 13 September 2026.

1438

ELO

400K

Context

$1.75

per 1M tokens

#41Open Source

Qwen3.8 27B

Alibaba

A 27-billion-parameter dense multimodal model under Apache 2.0, released 14 August 2026. Small enough to run yourself, takes text, images and video, and stretches from 262K to 1M tokens of context. Prices checked 1 September 2026. Arena text score 1437 (qwen3.8-27b, 10,697 votes), leaderboard of 13 September 2026.

1437

ELO

262K

Context

$0.50

per 1M tokens

#42Proprietary

o3

OpenAI

Advanced reasoning model succeeding o1, with significantly improved math and coding performance at reduced pricing. Arena text score 1432 (o3-2025-04-16, 58,579 votes), leaderboard of 13 September 2026.

1432

ELO

200K

Context

$10.00

per 1M tokens

#43Proprietary

Gemini 3.1 Flash-Lite

Google

Ultra-low latency model designed for sub-second responses at scale. Google's cheapest frontier-adjacent model. $0.25 input ($0.50 for audio) and $1.50 output per million tokens; prices checked 21 September 2026. Arena text score 1432 (gemini-3.1-flash-lite-preview, 60,405 votes), leaderboard of 13 September 2026.

1432

ELO

1M

Context

$0.25

per 1M tokens

#44Proprietary

Claude Opus 4

Anthropic

Anthropic's first Opus 4 generation model with extended thinking capabilities and strong agentic coding performance. Arena text score 1426 (claude-opus-4-20250514-thinking-16k, 35,735 votes), leaderboard of 13 September 2026.

1426

ELO

200K

Context

$5.00

per 1M tokens

#45Open Source

DeepSeek V3.2

DeepSeek

Updated open-source model with near-frontier capability at 1/20th the cost of GPT-5.5. Arena text score 1425 (deepseek-v3.2, 46,458 votes), leaderboard of 13 September 2026.

1425

ELO

128K

Context

$0.27

per 1M tokens

#46Open Source

DeepSeek R1

DeepSeek

Open-source reasoning model matching o1 performance with 97.3% MATH-500, disrupted the AI industry with its efficiency. Arena text score 1421 (deepseek-r1-0528, 18,091 votes), leaderboard of 13 September 2026.

1421

ELO

64K

Context

$0.55

per 1M tokens

#47Proprietary

Claude Haiku 4.5

Anthropic

Anthropic's fastest model in the 4.5 generation, offering near-Sonnet quality at Haiku-tier speed and pricing. Arena text score 1415 (claude-haiku-4-5-20251001, 129,278 votes), leaderboard of 13 September 2026.

1415

ELO

200K

Context

$1.00

per 1M tokens

#48Open Source

Mistral Large 3

Mistral AI

Open-weight Apache 2.0 MoE flagship from Mistral 3 generation with strong multilingual and multimodal performance. Arena text score 1413 (mistral-large-3, 69,028 votes), leaderboard of 13 September 2026.

1413

ELO

128K

Context

$2.00

per 1M tokens

#49Proprietary

Grok-3

xAI

Trained on xAI's Colossus supercluster, top-tier math reasoning with 93.3% AIME 2025 score. Arena text score 1412 (grok-3-preview-02-24, 32,414 votes), leaderboard of 13 September 2026.

1412

ELO

131K

Context

$3.00

per 1M tokens

#50Proprietary

Gemini 2.5 Flash

Google

Fast reasoning model with excellent cost efficiency, balancing speed and intelligence for high-throughput applications. Arena text score 1410 (gemini-2.5-flash, 122,732 votes), leaderboard of 13 September 2026.

1410

ELO

1M

Context

$0.30

per 1M tokens

#51Proprietary

o1

OpenAI

OpenAI's first reasoning model that uses chain-of-thought to solve complex math, science, and coding problems. Arena text score 1402 (o1-2024-12-17, 27,807 votes), leaderboard of 13 September 2026.

1402

ELO

200K

Context

$15.00

per 1M tokens

#52Proprietary

GPT-5.4 Nano

OpenAI

OpenAI's smallest GPT-5.4-class model, optimized for ultra-cheap classification, extraction, and lightweight agent sub-tasks. Arena text score 1402 (gpt-5.4-nano-high, 58,424 votes), leaderboard of 13 September 2026.

1402

ELO

400K

Context

$0.20

per 1M tokens

#53Proprietary

Claude Sonnet 4

Anthropic

Balanced mid-tier model in the Claude 4 generation, offering strong coding and reasoning at competitive pricing. Arena text score 1401 (claude-sonnet-4-20250514-thinking-32k, 33,944 votes), leaderboard of 13 September 2026.

1401

ELO

200K

Context

$3.00

per 1M tokens

#54Open Source

DeepSeek V3

DeepSeek

Chinese open-source MoE model rivaling GPT-4o at a fraction of the cost, trained for under $6M causing industry shock. Arena text score 1396 (deepseek-v3-0324, 44,787 votes), leaderboard of 13 September 2026.

1396

ELO

128K

Context

$0.14

per 1M tokens

#55Proprietary

Claude 3.5 Sonnet (Oct 2024)

Anthropic

Updated Sonnet with computer use capability and improved coding (93.7% HumanEval), the most popular coding model of late 2024. Arena text score 1374 (claude-3-5-sonnet-20241022, 87,694 votes), leaderboard of 13 September 2026.

1374

ELO

200K

Context

$3.00

per 1M tokens

#56Proprietary

o3-mini

OpenAI

Cost-efficient reasoning model with adjustable effort levels (low/medium/high), matching o1 at medium on STEM tasks. Arena text score 1363 (o3-mini-high, 18,589 votes), leaderboard of 13 September 2026.

1363

ELO

200K

Context

$1.10

per 1M tokens

#57Proprietary

Gemini 2.0 Flash

Google

Ultra-fast and affordable multimodal model with native tool use, image/audio generation, and 1M token context. Arena text score 1360 (gemini-2.0-flash-001, 43,351 votes), leaderboard of 13 September 2026.

1360

ELO

1M

Context

$0.10

per 1M tokens

#58Proprietary

Gemini 1.5 Pro

Google

Google's first million-token context model (up to 2M), excelling at long-document understanding and multimodal tasks. Arena text score 1351 (gemini-1.5-pro-002, 55,606 votes), leaderboard of 13 September 2026.

1351

ELO

2M

Context

$1.25

per 1M tokens

#59Proprietary

GPT-4o

OpenAI

OpenAI's flagship multimodal model with native text, vision, and audio capabilities, offering strong all-around performance. Arena text score 1346 (gpt-4o-2024-05-13, 112,881 votes), leaderboard of 13 September 2026.

1346

ELO

128K

Context

$2.50

per 1M tokens

#60Proprietary

Grok-2

xAI

xAI's second-gen model with real-time X/Twitter data access, competitive with GPT-4o on standard benchmarks. Arena text score 1336 (grok-2-2024-08-13, 63,498 votes), leaderboard of 13 September 2026.

1336

ELO

128K

Context

$2.00

per 1M tokens

#61Open Source

Llama 3.1 405B

Meta

Largest open-source model at release, competitive with GPT-4o and Claude 3.5 Sonnet across most benchmarks. Arena text score 1335 (llama-3.1-405b-instruct-bf16, 41,375 votes), leaderboard of 13 September 2026.

1335

ELO

128K

Context

$0.80

per 1M tokens

#62Open Source

Llama 4 Maverick

Meta

Higher-capability open-weight Llama 4 model in Meta's multimodal MoE generation, available for download via llama.com. Arena text score 1327 (llama-4-maverick-17b-128e-instruct, 39,359 votes), leaderboard of 13 September 2026.

1327

ELO

N/A

Context

per 1M tokens

#63Proprietary

Claude 3 Opus

Anthropic

Anthropic's original flagship model, excelling at complex analysis and nuanced writing with strong safety alignment. Arena text score 1322 (claude-3-opus-20240229, 194,909 votes), leaderboard of 13 September 2026.

1322

ELO

200K

Context

$15.00

per 1M tokens

#64Open Source

Llama 4 Scout

Meta

Open-weight, natively multimodal Llama 4 model designed for efficient deployment and long-context workloads. Arena text score 1321 (llama-4-scout-17b-16e-instruct, 29,740 votes), leaderboard of 13 September 2026.

1321

ELO

N/A

Context

per 1M tokens

#65Proprietary

GPT-4o Mini

OpenAI

Cost-efficient small model replacing GPT-3.5 Turbo, offering strong performance at a fraction of GPT-4o's cost. Arena text score 1318 (gpt-4o-mini-2024-07-18, 68,697 votes), leaderboard of 13 September 2026.

1318

ELO

128K

Context

$0.15

per 1M tokens

#66Proprietary

Mistral Large 2

Mistral AI

Mistral's flagship with 123B parameters, multilingual in 80+ languages, and strong code generation (92% HumanEval). Arena text score 1314 (mistral-large-2407, 45,459 votes), leaderboard of 13 September 2026.

1314

ELO

128K

Context

$2.00

per 1M tokens

#67Open Source

Qwen 2.5 72B Instruct

Alibaba

Alibaba's leading open-source model with strong multilingual and coding capabilities, competitive with Llama 3.1 70B. Arena text score 1303 (qwen2.5-72b-instruct, 39,406 votes), leaderboard of 13 September 2026.

1303

ELO

128K

Context

$0.30

per 1M tokens

#68Open Source

Llama 3.1 70B

Meta

Strong mid-size open-source model offering excellent performance-to-cost ratio for self-hosted deployments. Arena text score 1293 (llama-3.1-70b-instruct, 55,240 votes), leaderboard of 13 September 2026.

1293

ELO

128K

Context

$0.35

per 1M tokens

#69Proprietary

Claude 3 Haiku

Anthropic

Anthropic's fastest and most affordable model, designed for near-instant responses on simple queries and classification. Arena text score 1261 (claude-3-haiku-20240307, 117,701 votes), leaderboard of 13 September 2026.

1261

ELO

200K

Context

$0.25

per 1M tokens

#70Open Source

Llama 3.1 8B

Meta

Compact open-source model suitable for on-device and edge deployments with surprisingly strong capabilities for its size. Arena text score 1211 (llama-3.1-8b-instruct, 49,605 votes), leaderboard of 13 September 2026.

1211

ELO

128K

Context

$0.05

per 1M tokens

#71Proprietary

Claude Opus 5.5

Anthropic

Anthropic's new default flagship, released 22 September 2026, priced 20% below Opus 5 at $4/$20 with cache reads down 60% to $0.20 per million. 1M-token context, 128K output as standard and 300K behind a beta header. Tops Terminal-Bench 4.0 at 66.4% and CursorBench 4.0 at 57.8%, trails GPT-6 Astra on AutomationBench and Terminal-Bench Science; Anthropic notes the margins sit close to the noise. The advertised 40% cost saving compares Opus 5.5 at its default medium effort against Opus 5 at high, and at the same effort it thinks more per turn. Ships four breaking API changes, including thinking that cannot be disabled and a new computer-use toolset. Not yet listed on the Arena text leaderboard as of 13 September 2026. Prices checked 23 September 2026.

-

ELO

1M

Context

$4.00

per 1M tokens

#72Proprietary

GPT-6 Sol

OpenAI

Mid-tier of the GPT-6 family, released 22 September 2026 at half the price of GPT-5.6 Sol, and OpenAI states this is a permanent price rather than a promotion. Cached input $0.20 per million. 1.05M-token context, 128K output, knowledge cutoff 20 April 2026, reasoning effort from none to max with medium as default. On OpenAI's AutomationBench it completed 33.2% of agent tasks at $0.27 each, beating the far pricier GPT-6 Astra on both score and cost per task, and OpenAI says it makes about half as many factual mistakes as GPT-5.6 Sol. Not yet listed on the Arena text leaderboard as of 13 September 2026. Prices checked 23 September 2026.

-

ELO

1.05M

Context

$2.00

per 1M tokens

#73Proprietary

GPT-6 Luna

OpenAI

The cheap, high-volume tier of the GPT-6 family, released 22 September 2026 at $0.10/$0.50, half the input and 58% below the output price of GPT-5.6 Luna, and permanent rather than promotional. Cached input $0.01 per million. 1.05M-token context, 128K output, knowledge cutoff 18 May 2026. Scores 66.6% on DeepSWE 1.1, and OpenAI says that at higher effort it matches GPT-5.6 Sol's factual reliability at roughly one hundredth of the task cost. Reaches Free and Go users in the ChatGPT desktop app. Not yet listed on the Arena text leaderboard as of 13 September 2026. Prices checked 23 September 2026.

-

ELO

1.05M

Context

$0.10

per 1M tokens

#74Proprietary

Claude Mythos 5.1

Anthropic

The same model as Fable 5.1 without the production safeguards, released the same day and priced identically. Not generally available: access runs through restricted programmes for vetted cybersecurity and life-sciences organisations that need capabilities those safeguards normally constrain. Prices checked 1 September 2026. Not listed on the Arena text leaderboard as of 13 September 2026.

-

ELO

1M

Context

$10.00

per 1M tokens

#75Open Source

DeepSeek V4.1 Flash

DeepSeek

DeepSeek's September 10, 2026 replacement for V4 Flash, released as open weights under the MIT licence. 1M-token context and up to 384K tokens of output, with a much smaller KV cache that makes long cached prompts very cheap. Listed prices are peak-hour rates (01:00-04:00 and 06:00-10:00 UTC on weekdays); off-peak everything costs half, $0.15 input and $0.60 output, and cache hits are $0.003 off-peak. Not yet listed on the Arena text leaderboard as of 13 September 2026. Prices checked 21 September 2026.

-

ELO

1M

Context

$0.30

per 1M tokens

#76Proprietary

Seed 2.1 Turbo

ByteDance

ByteDance's cheap production workhorse, released in August 2026. Output length matches the context window, so it can write as much as it reads. Prices checked 1 September 2026. Not listed on the Arena text leaderboard as of 13 September 2026.

-

ELO

262K

Context

$0.50

per 1M tokens

#77Open Source

Qwen3.8 Flash

Alibaba

Released the same day as GLM 5.3 Flash, 26 August 2026, at almost the same price. Native context is 262K tokens and extends to 1M with YaRN. Prices checked 1 September 2026. Not listed on the Arena text leaderboard as of 13 September 2026.

-

ELO

262K

Context

$0.16

per 1M tokens

#78Proprietary

GPT-5.5 Pro

OpenAI

OpenAI's most capable model as of May 2026. 52.5% fewer hallucinations than GPT-5.4. Enhanced personalization with conversation memory and Gmail integration. Not listed on the Arena text leaderboard as of 13 September 2026.

-

ELO

256K

Context

$5.00

per 1M tokens