The model matrix
What do you need a model to do?
Say what the job is and get the model that fits it. Or browse them all by how smart, how fast and how affordable they are.
Or browse them yourself
Hold
StandardCost on a shared benchmark workload, thinking included.
Holding price at Standard: the chart compares the 7 models at that level on the other two.
Selected model
Claude Haiku 5.5
by Anthropic
- Smarts Smart
- Speed Instant
- Price Bargain
Reads text and images, files. Thinks before answering, as much or as little as you ask.
All models that fit
| Model | Smarts | Speed | Price | Thinks | Reads |
|---|---|---|---|---|---|
| Claude Opus 5.5Anthropic | Expert | Steady | Premium | Yes | images, files |
| GPT-6 AstraOpenAI | Expert | Unhurried | Premium | Yes | files, images |
| GPT-6.1 SolOpenAI | Expert | Unhurried | Budget | Yes | files, images |
| Claude Sonnet 5.5Anthropic | Expert | Quick | Premium | Yes | images, files |
| GPT-6.1 TerraOpenAI | Expert | Steady | Budget | Yes | files, images |
| Gemini 4 ArgonGoogle | Expert | Unhurried | Premium | Yes | audio, files, images, video |
| Claude Fable 5.1Anthropic | Expert | Unhurried | Premium | Yes | images, files |
| Kimi K3.1 ProKimi | Expert | Unhurried | Standard | Yes | images, video |
| Muse Spark 1.3Meta | Smart | Quick | Standard | Yes | images, video, files |
| Claude Haiku 5.5Anthropic | Smart | Instant | Bargain | Yes | images, files |
| Grok 4.7SpaceXAI | Smart | Unhurried | Standard | Yes | images, files |
| Qwen3.8 Max (0902)Alibaba | Smart | Unhurried | Premium | Yes | images, video |
| GLM 5.3Z AI | Smart | Steady | Standard | Yes | text only |
| Gemini 3.8 FlashGoogle | Smart | Steady | Premium | Yes | images, video, files, audio |
| MiMo-V2.6-ProXiaomi | Smart | Unhurried | Bargain | Yes | images, video, audio |
| MiniMax M3.1 FlashMiniMax | Capable | Quick | Budget | Yes | images, video |
| Step 5 PreviewStepFun | Capable | Quick | Standard | Yes | images, video |
| Solar Pro 4Upstage | Capable | Steady | unmeasured | Yes | text only |
| GLM 5.3 FlashZ AI | Capable | Quick | Budget | Yes | images, video |
| GPT-6 LunaOpenAI | Capable | Steady | Bargain | Yes | files, images |
| Mistral Large 4Mistral | Capable | Quick | unmeasured | Yes | images |
| MiniMax M3MiniMax | Capable | Steady | Budget | Yes | images, video |
| Nemotron 3.5 Super VLNVIDIA | Basic | Instant | Budget | Yes | images, video |
| Inkling SmallThinking Machines | Basic | Instant | Bargain | Yes | images, audio |
| InklingThinking Machines | Basic | Instant | Standard | Yes | images, audio |
| Nemotron 3 UltraNVIDIA | Basic | Quick | Standard | Yes | text only |
| Gemini 3.5 Flash LiteGoogle | Basic | Instant | Budget | Yes | images, video, files, audio |
| Command A+Cohere | Basic | Instant | Bargain | Yes | images |
| Nemotron 3.5 LightningNVIDIA | Basic | Instant | Bargain | Yes | text only |
| Mistral Small 4Mistral | Basic | Instant | Bargain | Yes | images |
How the levels are made
- Each model is ranked against the others in the matrix on each dimension, and the ranking is cut into four levels. "Expert" means the top quarter of the models offered, not a fixed score.
- Smarts comes from Artificial Analysis's intelligence index, republished by OpenRouter and cross-checked against Artificial Analysis's own listing.
- Speed is the median output rate of OpenRouter's live endpoints over the last half hour, or Artificial Analysis's measurement when the endpoints report none.
- Price compares the cost of a shared Artificial Analysis benchmark workload, including reasoning tokens, at its provider prices. It is not your agents.host bill. Missing measurements are marked unmeasured. Platform usage is shown separately and does not change the benchmark ranking.
On agents.host you choose your agent's model on this same chart, or leave it on Auto and let the platform choose per conversation.
Build an agent