Skip to main content
nexeai
Rankings

Which language models are actually used, and what for

The leading models for each task, in the order OpenRouter ranks them from the traffic going through its gateway. No score of ours: we show that ranking as it is, without re-sorting it.

Book a call

By task

The leading models for each task, in the order OpenRouter ranks them. The figure on the right is the input price, in dollars per million tokens.

Best LLMs for programming

  1. 1.GLM 5.3 Flashby Z.ai0.15
  2. 2.DeepSeek V4.1 Flashby DeepSeek0.14
  3. 3.MiMo-V2.5by Xiaomi0.14
  4. 4.DeepSeek V4 Flash 0731by DeepSeek0.03
  5. 5.Hy4 previewby Tencent0.83
  6. 6.Muse Spark 1.3 Contributorby Meta0.10
  7. 7.GLM 5.3by Z.ai1.40
  8. 8.GPT-5.6 Lunaby OpenAI0.20

Best LLMs for technical work

  1. 1.DeepSeek V4.1 Flashby DeepSeek0.14
  2. 2.GLM 5.3 Flashby Z.ai0.15
  3. 3.DeepSeek V4 Flash 0731by DeepSeek0.03
  4. 4.Gemini 3.8 Flashby Google0.75
  5. 5.Hy4 previewby Tencent0.83
  6. 6.GPT-5.6 Lunaby OpenAI0.20
  7. 7.Solar Pro 4by Upstage0.09
  8. 8.DeepSeek V4 Flash 0423by DeepSeek0.09

Best LLMs for science

  1. 1.GLM 5.3 Flashby Z.ai0.15
  2. 2.DeepSeek V4 Flash 0731by DeepSeek0.03
  3. 3.DeepSeek V4.1 Flashby DeepSeek0.14
  4. 4.GPT-5.6 Lunaby OpenAI0.20
  5. 5.Muse Spark 1.3 Contributorby Meta0.10
  6. 6.DeepSeek V4 Flash 0423by DeepSeek0.09
  7. 7.Solar Pro 4by Upstage0.09
  8. 8.Hy4 previewby Tencent0.83

Best LLMs for academic research

  1. 1.DeepSeek V4 Flash 0731by DeepSeek0.03
  2. 2.DeepSeek V4.1 Flashby DeepSeek0.14
  3. 3.GLM 5.3 Flashby Z.ai0.15
  4. 4.GPT-5.6 Lunaby OpenAI0.20
  5. 5.DeepSeek V4 Flash 0423by DeepSeek0.09
  6. 6.Hy4 previewby Tencent0.83
  7. 7.GPT-5.4 Nanoby OpenAI0.20
  8. 8.GLM 5.2by Z.ai0.65

Best LLMs for marketing

  1. 1.DeepSeek V4.1 Flashby DeepSeek0.14
  2. 2.GPT-5.6 Lunaby OpenAI0.20
  3. 3.GLM 5.3 Flashby Z.ai0.15
  4. 4.DeepSeek V4 Flash 0731by DeepSeek0.03
  5. 5.DeepSeek V4 Flash 0423by DeepSeek0.09
  6. 6.Gemma 4 26B A4B by Google0.09
  7. 7.Hy4 previewby Tencent0.83
  8. 8.Claude Sonnet 4.6by Anthropic3.00

Best LLMs for search engine optimisation

  1. 1.Qwen3.7 Flashby Qwen0.03
  2. 2.GPT-5.6 Solby OpenAI2.00
  3. 3.GLM 5.3 Flashby Z.ai0.15
  4. 4.DeepSeek V4 Flash 0731by DeepSeek0.03
  5. 5.GPT-5.2by OpenAI1.75
  6. 6.gpt-oss-120bby OpenAI0.15
  7. 7.DeepSeek V4 Flash 0423by DeepSeek0.09
  8. 8.DeepSeek V4.1 Flashby DeepSeek0.14

Best LLMs for translation

  1. 1.DeepSeek V4 Flash 0731by DeepSeek0.03
  2. 2.GPT-5.6 Lunaby OpenAI0.20
  3. 3.DeepSeek V4 Flash 0423by DeepSeek0.09
  4. 4.GLM 5.3 Flashby Z.ai0.15
  5. 5.DeepSeek V4.1 Flashby DeepSeek0.14
  6. 6.Gemini 2.5 Flash Liteby Google0.10
  7. 7.Gemini 3.1 Flash Liteby Google0.25
  8. 8.GPT-5.4 Nanoby OpenAI0.20

Best LLMs for finance

  1. 1.DeepSeek V4.1 Flashby DeepSeek0.14
  2. 2.DeepSeek V4 Flash 0731by DeepSeek0.03
  3. 3.GLM 5.3 Flashby Z.ai0.15
  4. 4.GLM 5.2by Z.ai0.65
  5. 5.GPT-5.6 Lunaby OpenAI0.20
  6. 6.Claude Sonnet 5by Anthropic2.00
  7. 7.DeepSeek V4 Flash 0423by DeepSeek0.09
  8. 8.Claude Opus 5by Anthropic5.00

Best LLMs for healthcare

  1. 1.GLM 5.3 Flashby Z.ai0.15
  2. 2.DeepSeek V4.1 Flashby DeepSeek0.14
  3. 3.GPT-5.6 Lunaby OpenAI0.20
  4. 4.DeepSeek V4 Flash 0731by DeepSeek0.03
  5. 5.GPT-5 Nanoby OpenAI0.05
  6. 6.DeepSeek V4 Flash 0423by DeepSeek0.09
  7. 7.Qwen3.8 Max (0902)by Qwen2.00
  8. 8.Gemini 3.8 Flashby Google0.75

Best LLMs for roleplay

  1. 1.DeepSeek V4 Flash 0423by DeepSeek0.09
  2. 2.DeepSeek V4 Flash 0731by DeepSeek0.03
  3. 3.DeepSeek V3.2by DeepSeek0.27
  4. 4.GLM 5.3 Flashby Z.ai0.15
  5. 5.MiMo-V2.5by Xiaomi0.14
  6. 6.MiMo-V2.5-Proby Xiaomi0.44
  7. 7.Gemini 3 Flash Previewby Google0.50
  8. 8.DeepSeek V4.1 Flashby DeepSeek0.14

Best LLMs for general knowledge

  1. 1.DeepSeek V4.1 Flashby DeepSeek0.14
  2. 2.DeepSeek V4 Flash 0423by DeepSeek0.09
  3. 3.GPT-5.6 Lunaby OpenAI0.20
  4. 4.GLM 5.3 Flashby Z.ai0.15
  5. 5.DeepSeek V4 Pro 0813by DeepSeek0.46
  6. 6.Gemini 3.8 Flashby Google0.75
  7. 7.GLM 5.2by Z.ai0.65
  8. 8.DeepSeek V4 Flash 0731by DeepSeek0.03

By capability

The same catalogue, sorted this time on the Artificial Analysis intelligence index, which OpenRouter publishes for some of the models. Compare it with the lists above: the two rankings do not always match.

  1. 1.Claude Opus 5.5by Anthropic57.6
  2. 2.Claude Fable 5.1by Anthropic53.4
  3. 3.GPT-6 Astraby OpenAI52.7
  4. 4.Claude Opus 5by Anthropic50.8
  5. 5.Claude Fable 5by Anthropic49.6
  6. 6.GPT-6 Solby OpenAI47.5
  7. 7.GPT-5.6 Solby OpenAI47
  8. 8.Grok 4.7by SpaceXAI46.4
  9. 9.MiMo-V2.6-Proby Xiaomi46.3
  10. 10.Qwen3.8 Max (0902)by Qwen45.4

Artificial Analysis index, published by OpenRouter

By context window

How much text a model holds in view at once. It decides the size of the file you can hand it in one go: a contract, a set of accounts, a codebase.

  1. 1.Grok 4.20 Multi-Agentby SpaceXAI2 M
  2. 2.Grok 4.20by SpaceXAI2 M
  3. 3.GLM Flash Latestby Z.ai1.3 M
  4. 4.GLM 5.3 Flashby Z.ai1.3 M
  5. 5.GLM Latestby Z.ai1.3 M
  6. 6.GLM 5.3by Z.ai1.3 M
  7. 7.DeepSeek V4 Flash Latestby DeepSeek1.3 M
  8. 8.DeepSeek V4 Flash 0731by DeepSeek1.3 M
  9. 9.Llama 4 Scoutby Meta1.3 M
  10. 10.GPT-6 Luna Proby OpenAI1.1 M

Context window, in tokens

What this ranking measures, and what it does not

What is counted
Traffic routed through the OpenRouter API. We show the models in the order its API returns for each task, without re-sorting; on its rankings page, OpenRouter states that tasks are ranked by share of spend. Its datasets only cover public traffic: private models, private endpoints and zero-data-retention traffic are excluded at the source.
What it does not say
Neither accuracy nor reasoning quality. Rank also follows the price and verbosity of models, not only their results. That is what the “By capability” section is for.
The scope
Traffic through OpenRouter, not the whole market. A model used heavily on its own vendor's API appears lower here than it deserves.
Freshness
The page re-reads itself once a day. The date shown is the date of the reading, not of the deployment.

Source: OpenRouter (openrouter.ai/rankings), retrieved on 24 September 2026. Rankings licensed under CC BY 4.0.

Picking a model is the last question, not the first

What decides an agent is the work it is given, the data it touches and where it runs. The model itself changes in one line. Let's talk about the rest.

Book a call