How good, how fast, what it costs to pick wrong.
Almost every board on this page orders models by a figure someone else measured and published: quality by Artificial Analysis and Design Arena, speed by OpenRouter timing the traffic it routes, and seller spread by what the companies selling a model ask for it, which ranks nothing about the model, only how much the choice of seller is worth on it. The usage board is the exception and says so: the tokens each model ran through OpenRouter, as OpenRouter publishes them, which is adoption on one router rather than a measure of the model.
58 models scored by Artificial Analysis · 132 ranked by Design Arena · 324 timed by OpenRouter · 186 sold by more than one company · 84 ranked by OpenRouter usage · 359 models in the catalog·retrieved sha256 · 0bb53a77713f
Claude Fable 5.2 leads at 64.6; 58 models scored.
- 01Claude Fable 5.2Anthropic64.6
- 02Gemini 4 CarbonGoogle60.0
- 03Claude Opus 5.5Anthropic59.1
- 04GPT-6 AstraOpenAI56.9
- 05GPT-6.1 SolOpenAI54.6
- 06Claude Sonnet 5.5Anthropic54.5
- 07Gemini 4 HydrogenGoogle53.9
- 08GPT-6.1 TerraOpenAI51.3
- 09Gemini 4 ArgonGoogle50.4
- 10Claude Fable 5.1Anthropic50.0
- 11Kimi K3.1 ProMoonshotAI46.5
- 12Claude Opus 5Anthropic44.2
- 13GPT-6 SolOpenAI44.2
- 14Muse WonderMeta43.6
- 15Muse Spark 1.3Meta36.6
- 16Claude Haiku 5.5Anthropic36.5
- 17Grok 4.7SpaceXAI35.5
- 18Hy4Tencent34.6
- 19Qwen3.8 Max (0902)Qwen33.7
- 20GPT-5.6 TerraOpenAI32.3
- 21GLM 5.3Z.ai31.7
- 22Grok 4.6SpaceXAI31.0
- 23Gemini 3.8 FlashGoogle30.8
- 24Inkling 2 SmallThinking Machines30.6
- 25MiMo-V2.6-ProXiaomi29.9
- 26Ling 3.1 FlashinclusionAI29.6
The top 26 of 58, 64.6 down to 29.6 on the Intelligence Index. Every column stands on zero on the index's own 0 to 100 scale, so a column half as tall is half the score, and the maker's mark hangs under each. 32 more are scored and ranked on the intelligence leaderboard.
15 models have held the Coding Index ceiling since May 2023; Claude Fable 5.1 holds it at 80.7.
- a scored model, in its maker's tone
- raised the ceiling, named
- the running best
- 01GPT-413.1 · listed 28 May 2023
- 02GPT-4 Turbo (batch)21.5 · listed 9 Apr 2024
- 03GPT-4o (2024-05-13)24.2 · listed 13 May 2024
- 04Gemini 2.5 Pro33.3 · listed 17 Jun 2025
- 05GPT-537.8 · listed 7 Aug 2025
- 06DeepSeek V3.1 Terminus43.5 · listed 22 Sep 2025
- 07Claude Sonnet 4.552.1 · listed 29 Sep 2025
- 08Claude Sonnet 4.663.0 · listed 17 Feb 2026
- 09Gemini 3.1 Pro Preview68.8 · listed 19 Feb 2026
- 10GPT-5.471.1 · listed 5 Mar 2026
- 11Claude Opus 4.773.6 · listed 16 Apr 2026
- 12GPT-5.574.9 · listed 24 Apr 2026
- 13Claude Fable 576.5 · listed 9 Jun 2026
- 14GPT-5.6 Sol78.3 · listed 9 Jul 2026
- 15Claude Fable 5.180.7 · listed 1 Sep 2026
Each dot is one of the 118 scored models, at the date OpenRouter listed it, which is not always the date the provider released it, filled in its maker's tone as the columns above are. The staircase is the running best, and the models that raised it are named on the figure and numbered in the key. The axis runs from zero to 100, the first round tick over the ceiling, rather than to the index's 100, because nothing here is a length. Scores are the Coding Index as measured today; the catalog does not carry past scores, so this lays the current measurement over the listing dates rather than replaying history.
Claude Fable 5.1, 80.7.
- 01Claude Fable 5.1Anthropic80.7
- 02GPT-5.6 SolOpenAI78.3
- 03Claude Opus 5Anthropic78.0
- 04GPT-6 AstraOpenAI76.9
- 05GPT-5.6 TerraOpenAI76.7
- 06Claude Fable 5Anthropic76.5
- 07Gemini 3.8 FlashGoogle76.3
- 08Qwen3.8 Max (0902)Qwen76.2
- 09Gemini 3.7 FlashGoogle76.1
- 10Grok 4.6SpaceXAI75.9
- 11Muse Spark 1.3Meta75.8
- 12GPT-5.5OpenAI74.9
- 13GLM 5.3Z.ai74.8
- 14Claude Opus 4.8Anthropic74.3
- 15Claude Opus 4.7Anthropic73.6
- 16Grok 4.5SpaceXAI72.4
- 17Muse Spark 1.2Meta72.2
- 18Kimi K3MoonshotAI72.0
- 19Qwen3.8 2.4T A95BQwen71.9
- 20Claude Sonnet 5Anthropic71.5
Claude Fable 5.1, 57.2.
- 01Claude Fable 5.1Anthropic57.2
- 02Claude Opus 5Anthropic56.5
- 03Qwen3.8 Max (0902)Qwen56.0
- 04Muse Spark 1.3Meta55.5
- 05GLM 5.3Z.ai53.1
- 06Grok 4.6SpaceXAI52.5
- 07GPT-6 AstraOpenAI51.0
- 08GLM 5.3 FlashZ.ai50.9
- 09Claude Fable 5Anthropic50.7
- 10Qwen3.8 2.4T A95BQwen50.1
- 11DeepSeek V4 Flash Vision ExpDeepSeek47.5
- 12GPT-5.6 SolOpenAI47.4
- 13Qwen3.8 27BQwen45.8
- 14Claude Sonnet 5Anthropic43.6
- 15Muse Spark 1.2Meta43.2
- 16GPT-5.6 TerraOpenAI43.2
- 17GPT-5.6 LunaOpenAI42.1
- 18Claude Opus 4.8Anthropic41.9
- 19DeepSeek V4 Pro 0813DeepSeek41.3
- 20Grok 4.5SpaceXAI41.2
gpt-oss-safeguard-20b writes fastest at 438 tok/s; Granite 4.2 8B answers first at 148ms.
Output tokens per second, median.
- 01gpt-oss-safeguard-20bOpenAI438
- 02LFM2.5-24B-A2BLiquidAI432
- 03Beam 1.0 501B A23BReflection417
- 04LFM3-1.7BLiquidAI290
- 05Llama 3.1 8B InstructMeta Llama266
- 06Nemotron 3.5 LightningNVIDIA260
- 07gpt-oss-20bOpenAI249
- 08Nemotron 3 Nano 30B A3BNVIDIA231
- 09MiniMax M2.7MiniMax224
- 10Muse Spark OpenMeta224
- 11gpt-oss-120bOpenAI223
- 12Muse EmberMeta223
- 13Nemotron 3.5 Super VLNVIDIA214
- 14Nemotron 3 SuperNVIDIA204
- 15Inkling 2 SmallThinking Machines192
- 16Gemini 2.5 Flash LiteGoogle186
- 17Gemini 3.1 Flash Lite PreviewGoogle172
- 18Ministral 3 3B 2512Mistral167
- 19DeepSeek V4.1 FlashDeepSeek150
- 20Nova Micro 1.0Amazon148
Wait before the first token, median.
- 01Granite 4.2 8BIBM Granite148ms
- 02LFM3-1.7BLiquidAI152ms
- 03LFM2.5-24B-A2BLiquidAI158ms
- 04Phi 4Microsoft190ms
- 05Llama 3 8B LunarisSao10K204ms
- 06Codestral 2508Mistral211ms
- 07Mistral Small 3Mistral227ms
- 08gpt-oss-safeguard-20bOpenAI244ms
- 09Seed-2.0-MiniByteDance Seed263ms
- 10Nemotron 3.5 LightningNVIDIA275ms
- 11Beam 1.0 501B A23BReflection286ms
- 12Llama Guard 4 12BMeta Llama294ms
- 13Llama 3.1 8B InstructMeta Llama311ms
- 14Nemotron 3 Nano 30B A3BNVIDIA316ms
- 15Hermes 4 405BNous329ms
- 16Muse Spark OpenMeta337ms
- 17gpt-oss-120bOpenAI339ms
- 18Llama 4 ScoutMeta Llama355ms
- 19Qwen2.5 72B InstructQwen370ms
- 20Nova Micro 1.0Amazon371ms
The top 20 of 241, 438 down to 148 in output tokens per second. Every column stands on zero, so a column half as tall is half the rate. OpenRouter timed its own routed traffic over a 30-minute window and published the percentiles per model; 324 of 359 models carried a figure, and 241 of those cleared 250 requests and are ranked here. A model below that floor keeps its figure on its own page, where the request count sits beside it, but a good half hour over eleven requests is not a ranking. Unlike every other board on this page, these are not scores of a model: the same weights behind a different provider answer at a different speed, so a model's page carries every route separately. Read .
Ling 3.1 Flash: 439 index points per blended dollar.
Intelligence Index divided by the blended OpenRouter list price per million tokens, 3 input tokens to 1 output. 57 priced models have both figures. 1 scored listing is free and left out: a zero price makes the ratio undefined, not infinite.
Claude Fable 5.1 leads web apps in the agents arena at 1319 Elo.
| Position | Model | Elo | Win rate | Intelligence |
|---|---|---|---|---|
| 01 | Claude Fable 5.1 | 1319 | 58.2% | 50.0 |
| 02 | Claude Opus 5.5 | 1319 | 62.6% | 59.1 |
| 03 | Kimi K3.1 Pro | 1307 | 60.0% | 46.5 |
| 04 | Claude Sonnet 5.5 | 1297 | 58.7% | 54.5 |
| 05 | Muse Spark 1.3 | 1288 | 55.8% | 36.6 |
| 06 | GPT-6.1 Sol | 1287 | 57.4% | 54.6 |
| 07 | Gemini 4 Argon | 1283 | 47.5% | 50.4 |
| 08 | Kimi K3 | 1277 | 58.0% | 17.4 |
| 09 | MiMo-V2.6-Pro | 1272 | 58.2% | 29.9 |
| 10 | GLM 5.3 | 1270 | 54.2% | 31.7 |
| 11 | Kimi K2.6 | 1268 | 59.3% | not scored |
| 12 | Claude Opus 5 | 1248 | 53.3% | 44.2 |
| 13 | Claude Fable 5 | 1242 | 56.3% | not scored |
| 14 | Grok 4.6 | 1230 | 55.2% | 31.0 |
| 15 | Claude Sonnet 5 | 1222 | 51.8% | not scored |
| 16 | Gemini 3.8 Flash | 1216 | 45.5% | 30.8 |
| 17 | Claude Opus 4.8 | 1213 | 50.3% | not scored |
| 18 | Gemini 3.7 Flash | 1209 | 47.5% | not scored |
| 19 | Muse Spark 1.2 | 1204 | 50.0% | not scored |
| 20 | GLM 5.2 | 1203 | 55.6% | not scored |
| 21 | Grok 4.5 | 1198 | 53.0% | not scored |
| 22 | MiniMax M3 | 1190 | 47.7% | 18.9 |
| 23 | Qwen3.7 Max | 1189 | 48.3% | not scored |
| 24 | Muse Spark 1.1 | 1187 | 48.7% | not scored |
| 25 | Claude Opus 4.6 | 1186 | 55.6% | not scored |
47 models ranked for web apps in the agents arena; Elo from pairwise human votes at Design Arena, win rate the share of matchups won. Position is the place in this catalog; the author column and Design Arena's own rank, which counts models the catalog does not carry, are on the full board. 22 more on the full board.
4k to 2M: 34 window sizes across 359 models.
Each tick is one context window the catalog lists, on a logarithmic scale, with how many models carry it. Every window below opens to the models that carry it, widest first.
1.31M1 model
1.05M85 models
Beam 1.0 501B A23BDeepSeek V4 Flash 0423DeepSeek V4 Flash 0731DeepSeek V4 Flash Vision ExpDeepSeek V4 Pro 0423DeepSeek V4 Pro 0813DeepSeek V4.1 FlashDeepSeek V4.1 Flash 1017Ember-1GLM 5.2GLM 5.3GLM 5.3 FlashGLM 5.3 FlashXGLM 5.3 FlashX LegalGPT-4.1GPT-4.1 MiniGPT-4.1 NanoGPT-5.4GPT-5.4 ProGPT-5.5GPT-5.5 ProGPT-5.6 LunaGPT-5.6 Luna ProGPT-5.6 SolGPT-5.6 Sol ProGPT-5.6 TerraGPT-5.6 Terra ProGPT-6 AstraGPT-6 Astra ProGPT-6 LunaGPT-6 Luna ProGPT-6 SolGPT-6 Sol ProGPT-6.1 SolGPT-6.1 Sol ProGPT-6.1 TerraGPT-6.1 Terra ProGemini 2.5 FlashGemini 2.5 Flash LiteGemini 2.5 ProGemini 2.5 Pro Preview 06-05Gemini 3 Flash PreviewGemini 3.1 Flash LiteGemini 3.1 Flash Lite PreviewGemini 3.1 Pro PreviewGemini 3.1 Pro Preview Custom ToolsGemini 3.5 FlashGemini 3.5 Flash LiteGemini 3.6 FlashGemini 3.7 FlashGemini 3.8 FlashGemini 4 ArgonGemini 4 CarbonGemini 4 HydrogenHy4Hy4 flashHy4 previewInkling 2 SmallKimi K3Kimi K3.1 ProLing 3.2 Flash (free)LongCat 2.0Lyria 3 Clip PreviewLyria 3 Pro PreviewMiMo-V2.6-FlashMiMo-V2.6-ProMiMo-V2.6-Pro-UltraSpeedMiniMax M3MiniMax M3.1 FlashMistral Large 4Muse EmberMuse Spark 1.1Muse Spark 1.2Muse Spark 1.2 ContributorMuse Spark 1.3Muse Spark 1.3 ContributorMuse Spark OpenMuse WonderNemotron 3.5 LightningNemotron 3.5 Super VLPareto 26.10Pareto 26.10 PreviewParticle-1Qwen3.8 2.4T A95BRipple 65B Preview
1.04M1 model
1M43 models
Claude Fable 5Claude Fable 5.1Claude Fable 5.2Claude Haiku 5.5Claude Opus 4.6Claude Opus 4.7Claude Opus 4.8Claude Opus 5Claude Opus 5.5Claude Sonnet 4.5Claude Sonnet 4.6Claude Sonnet 5Claude Sonnet 5.5ExtractFugu MaxFugu UltraFugu Ultra v2FusionGLM 5.3 PrimeGrok 4.3Jev RouterLing 3.1 FlashMiMo-V2.5Nova 2 LiteNova Premier 1.0Qwen-PlusQwen3 Coder FlashQwen3.5 Plus 2026-02-15Qwen3.5 Plus 2026-04-20Qwen3.5-FlashQwen3.6 FlashQwen3.6 PlusQwen3.7 FlashQwen3.7 MaxQwen3.7 PlusQwen3.8 FlashQwen3.8 Max (0902)Qwen3.8 Max PrimeQwen3.8 Omni FlashQwen4 MaxStep 5 PreviewSummarizeSwitchyard
512k1 model
300k2 models
272k1 model
262k56 models
Aion 3.5Aion 3.5 MiniDevstral 2 2512Flint (free)Gemma 4 26B A4BGemma 4 31BGlimpseHy3Hy3 previewKimi K2 0905Kimi K2 ThinkingKimi K2.5Kimi K2.6Kimi K2.7 CodeLFM3-1.7BLing 3.0 FlashLing 3.0 TinyMi:dm K 3.0 ProMinistral 3 14B 2512Ministral 3 8B 2512Mistral Large 3 2512Mistral Medium 3.5Mistral Small 4Morph V3 FastMorph V3 LargeMotif 3Nemotron 3 Nano 30B A3BNemotron 3 SuperNemotron 3 UltraNex-N2.5-MiniNex-N2.5-ProNex-N2.6Pareto 26.9Qwen3 235B A22B Instruct 2507Qwen3 30B A3B Instruct 2507Qwen3 Coder 30B A3B InstructQwen3 Coder 480B A35BQwen3 Coder NextQwen3 Next 80B A3B InstructQwen3 VL 235B A22B InstructQwen3 VL 30B A3B InstructQwen3 VL 8B InstructQwen3.5 397B A17BQwen3.5-122B-A10BQwen3.5-27BQwen3.5-35B-A3BQwen3.5-9BQwen3.6 27BQwen3.6 35B A3BSakana NamazuSeed 2.1 TurboSeed-2.0-LiteSeed-2.0-MiniStep 3.7 FlashTernary Bonsai 2 27BTernary Bonsai 2 9B
203k2 models
198k1 model
197k1 model
192k1 model
131k34 models
Aion-2.0Aion-3.0Aion-3.0-MiniCydonia 24B V4.1GLM 4.5GLM 4.5 AirGLM 4.6VGemini Omni 1.1 FlashGemini Omni 1.2 FlashGemma 3 12BGemma 3 27BGemma 3 4BGranite 4.2 8BHermes 4 405BKimi K2 0711Ling 3.0 Flash VLLlama 3.1 70B InstructLlama 3.1 8B InstructLlama 3.1 Euryale 70B v2.2Llama 3.2 3B InstructLlama 3.3 70B InstructMinistral 3 3B 2512Mistral Large 2407Mistral Medium 3Mistral Medium 3.1Mistral NemoMuse Glimmer 30BNano Banana 2 (Gemini 3.1 Flash Image)Nano Banana Pro (Gemini 3 Pro Image)Nemotron 3.5 Content SafetyQwen3 32Bgpt-oss-120bgpt-oss-20bgpt-oss-safeguard-20b
128k26 models
Body Builder (beta)Command R (08-2024)Command R+ (08-2024)Command R7B (12-2024)GPT AudioGPT Audio MiniGPT-4 Turbo (batch)GPT-4oGPT-4o (2024-05-13)GPT-4o (2024-08-06)GPT-4o (2024-11-20)GPT-4o-miniGPT-4o-mini (2024-07-18)GPT-5.2 ChatLFM2.5-24B-A2BMistral LargeMistral Small 3.1 24BNova Micro 1.0Qwen2.5 VL 72B InstructQwen3 235B A22B Thinking 2507Schematron V2 SmallSchematron V2 TurboSonar Deep ResearchSonar Reasoning ProUI-TARS 7BUncensored
127k1 model
60k1 model
41k2 models
37k1 model
32k2 models
16k3 models
6k1 model
4k1 model
21 models listed in Nov; 27 in the last thirty days.
- Qwen4 MaxQwennot scored
- Hy4 flashTencentnot scored
- Ling 3.2 Flash (free)inclusionAInot scored
- Gemini 4 HydrogenGoogleintelligence 53.9
- ERNIE-5.5-Flash-Preview (free)Baidunot scored
- ERNIE-Light-DT-1.0Baidunot scored
- Gemini 4 CarbonGoogleintelligence 60.0
- Nex-N2.6Nex AGInot scored
- Claude Fable 5.2Anthropicintelligence 64.6
- Muse EmberMetanot scored
- Inkling 2 SmallThinking Machinesintelligence 30.6
- LFM3-1.7BLiquidAInot scored
- Gemini Omni 1.2 FlashGooglenot scored
- Mi:dm K 3.0 ProKTnot scored
- Auto RouterOpenRouternot scored
- Auto CodeOpenRouternot scored
- Gemini Omni 1.1 FlashGooglenot scored
- Ripple 65B PreviewRicursivenot scored
- Hy4Tencentintelligence 34.6
- Motif 3Motif Technologiesnot scored
- GLM 5.3 FlashX LegalZ.ainot scored
- Particle-1Particle AInot scored
- GPT-6.1 TerraOpenAIintelligence 51.3
- GPT-6.1 Terra ProOpenAInot scored
- Kimi K3.1 ProMoonshotAIintelligence 46.5
- Flint (free)Recursivenot scored
- MiniMax M3.1 FlashMiniMaxintelligence 29.1
- DeepSeek V4.1 Flash 1017DeepSeekintelligence 28.6
- GlimpseEloriannot scored
- Ternary Bonsai 2 9BPrismMLnot scored
The 30 newest listings, by the date OpenRouter listed them, which is not always the date the provider released them.
98 of the 186 models sold by more than one company cost at least twice as much from the dearest seller as the cheapest; the median is 2x.
The same weights, sold by several companies at different prices: each bar is how much dearer the dearest seller is than the cheapest, per million input tokens. The 20 widest of the 186 are drawn, and 25 of them are 5x or wider. The bars are linear, so a bar twice as long is twice the ratio. Unlike every other board on this page, this ranks nothing about a model: two models with the same spread are not equally good, and a wide spread means only that which company serves the request is worth deciding. Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. Read .
DeepSeek V4.1 Flash 1017 ran 23.9% of the tokens OpenRouter counted over the 30 days to 20 Nov 2026, and the top 50 models of each day ran 94.4% of them.
- 01DeepSeek V4.1 Flash 1017DeepSeek · in the top 50 on 30 of 30 days23.9%231T
- 02Hy4 flashTencent · in the top 50 on 5 of 30 days9.6%92.7T
- 03DeepSeek V4.1 FlashDeepSeek · in the top 50 on 30 of 30 days7.2%69.3T
- 04GLM 5.3 FlashZ.ai · in the top 50 on 30 of 30 days5.9%56.8T
- 05GPT-6 LunaOpenAI · in the top 50 on 30 of 30 days3.8%37.0T
- 06Hy4Tencent · in the top 50 on 17 of 30 days3.8%36.7T
- 07Claude Opus 5.5Anthropic · in the top 50 on 30 of 30 days3.2%31.0T
- 08Step 5 PreviewStepFun · in the top 50 on 29 of 30 days2.7%25.7T
- 09MiniMax M3.1 FlashMiniMax · in the top 50 on 29 of 30 days1.9%18.4T
- 10Hy4 previewTencent · in the top 50 on 17 of 30 days1.8%17.3T
- 11GLM 5.3Z.ai · in the top 50 on 30 of 30 days1.8%17.2T
- 12GPT-6.1 SolOpenAI · in the top 50 on 30 of 30 days1.6%15.3T
- 13MiMo-V2.6-FlashXiaomi · in the top 50 on 30 of 30 days1.6%15.1T
- 14Nemotron 3 Ultra (free)NVIDIA · in the top 50 on 30 of 30 days1.5%14.9T
- 15DeepSeek V4 Flash 0731DeepSeek · in the top 50 on 30 of 30 days1.5%14.6T
- 16Kimi K3.1 ProMoonshotAI · in the top 50 on 29 of 30 days1.4%13.5T
- 17typesafe/jev-1.13-20260917typesafe · in the top 50 on 30 of 30 days1.3%12.8T
- 18Claude Haiku 5.5Anthropic · in the top 50 on 30 of 30 days1.3%12.1T
- 19Muse Spark 1.3 ContributorMeta · in the top 50 on 30 of 30 days1.2%12.0T
- 20GPT-5.6 LunaOpenAI · in the top 50 on 30 of 30 days1.2%11.2T
The apps sending the tokens, 12 of the 50 OpenRouter ranks
- 01Hermes Agent55.9T498M requests
- 02Claude Code45.9T440M requests
- 03Kilo Code25.6T243M requests
- 04pi20.4T194M requests
- 05Codex15.3T234M requests
- 06OpenClaw13.8T106M requests
- 07omp10.4T92.8M requests
- 08Strix9.69T89.1M requests
- 09DeepSeek Harness9.57T59.3M requests
- 10Portkey AI6.46T448M requests
- 11Freebuff4.05T36.0M requests
- 12Minds by Authentic Intelligence3.88T105M requests
Share of every token OpenRouter counted from 22 Oct 2026 to 20 Nov 2026, 967T in all, including the traffic of every model outside the top 50 of each day, which OpenRouter publishes as one row. A model ranks by its tokens over the whole window; the line beside it is its share of each day on its own scale from zero, and a gap is a day it was outside the top 50 rather than a day it ran nothing. Token totals are prompt plus completion tokens as each provider tokenizer counts them, so a count on one model does not compare directly with a count on another. They are traffic through OpenRouter only, not the whole market: a model sold mostly through its maker directly can be large and rank low here. A ranked model the catalog does not list is named by the id OpenRouter ranks it under, without a link. Source: OpenRouter (openrouter.ai/rankings), as of 2026-11-21T16:13:12.792Z. CC BY 4.0. Source: OpenRouter (openrouter.ai/apps), as of 2026-11-21T16:13:13.837Z. CC BY 4.0. Read .
How these rankings are measured.
Four plates, read left to right: what was measured, when it was read, what it does not establish, and where every figure came from.
- 01What is measured.
Three Artificial Analysis indexes, each 0 to 100, and Design Arena's Elo per board, exactly as OpenRouter's catalog carries them; OpenRouter's own timings of the traffic it routes; and what each company selling a model asks for it. The usage board is not a measurement: it is the tokens each model and app ran through OpenRouter, as OpenRouter counts them. Vitna re-ran none of them and adjusted nothing. A ranking counts one row per model; OpenRouter's batch and free listings of the same model are not ranked twice.
- 02Freshness.
The catalog was read 17 Nov 16:29 UTC and this page is rebuilt when a read changes a figure. The catalog does not carry the dates the indexes were run, so a score's own date is not printed here.
- 03What they do not establish.
Not quality from adoption: the usage board says how much a model ran through one router, and no other board uses it. Not Vitna's evaluation: the engine scores quality on your own representative cases, and none of that is here. A third-party index is a third party's opinion, dated.
- 04Where the figures come from.
Intelligence Index, Coding Index, Agentic Index from Artificial Analysis; Elo, win rate and rank from Design Arena; both as republished in OpenRouter's public model list, cited below with the sha256 of the document read. Throughput and first-token times from OpenRouter's own model pages, per-seller prices from its public endpoints API, and usage from its keyed rankings datasets under CC BY 4.0, each mirrored on its own schedule and dated where it is drawn.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Vitna ran none of these evaluations and adjusted none of these figures.
