Not a vendor chart. These are live calls through withConflux: the typical answer time, the slow one, how soon the first word arrives, and how fast the rest streams. A range, not just a rank.
| Model | Answer time range | Typical | Slow (1 in 20) | First token | Output speed | Calls | Compare |
|---|---|---|---|---|---|---|---|
Claude Fable 5.1 Anthropic · claude-fable-5-1 | Not enough calls yet (30 of 100) | - | - | - | - | 30 | Compare |
Claude Haiku 5.5 Anthropic · claude-haiku-5-5 | Not enough calls yet (3 of 100) | - | - | - | - | 3 | Compare |
GPT-6.1 Sol OpenAI · gpt-6.1-sol | Not enough calls yet (2 of 100) | - | - | - | - | 2 | Compare |
Claude Opus 5.5 Anthropic · claude-opus-5-5 | Not enough calls yet (1 of 100) | - | - | - | - | 1 | Compare |
GPT-6 Astra OpenAI · gpt-6-astra | Not enough calls yet (0 of 100) | - | - | - | - | 0 | Compare |
GPT-6 Luna OpenAI · gpt-6-luna | Not enough calls yet (0 of 100) | - | - | - | - | 0 | Compare |
GPT-6.1 Terra OpenAI · gpt-6.1-terra | Not enough calls yet (0 of 100) | - | - | - | - | 0 | Compare |
Claude Sonnet 5.5 Anthropic · claude-sonnet-5-5 | Not enough calls yet (0 of 100) | - | - | - | - | 0 | Compare |
The middle call: half are faster, half slower. The number to plan with.
How soon the first words appear when streaming. What a chat user feels.
Words per second after the first one. What a long answer feels like.
Put any two models side by side with the cost for real workloads, or upload your bill and see what it would cost here.