Time to first token and throughput, exactly as the runtime source published them.
Time to first token: latency to the first answer chunk (Artificial Analysis, via BenchLM). Reasoning models include thinking time, so values can run to tens or hundreds of seconds.
Economics
Source rates plus a labelled 75% input / 25% output blended cost.
1 of these rows is marked self-hosted by the source: open-weight models with no first-party hosted token rate. 1 of them shows a hosted rate matched from another provider; the rest report an unavailable price rather than zero.
Decision deltas
Each row comes from the pair comparison source. A null basis or average stays unavailable rather than being inferred from the capability table.
No pair comparison was published for this selection.
The comparison source only publishes deltas for pairs it has evaluated. The capability and price tables above remain source-exact.
ModelScale