Benchmarks
Live measurements of every model on Miavo, taken from real API traffic rather than a one-off test run: how often requests succeed, what they cost and how fast the first token arrives. Each score links to the full per-model breakdown.
Language models
Chat and reasoning models grouped by family, measured on every live request.
Media
Images, video and speech, timed end to end and priced by what each output actually cost.
How we measure
Speed and latency use successful requests only. Success rate leaves out client errors such as invalid requests, auth, balance and rate limits. A metric shows once it has at least 5 samples. Blended price weights input 3:1 over output. First token and output speed are recorded from October 31, 2026.
For usage-based views of the same models, see the model rankings and the full model list.