One API for chat, image and speech modelsOpenAI-compatibleNo per-request markupCredit never expiresRead the docs
Models · 4/6
01ChartLoading
Filters and reference
Cost / request ($)Writing Elo
02Side by side8,000 in · 500 out · no cache hits
Model1 request100 requestsvs. referenceWriting EloFirst tokenSpeedContext

Best marks the strongest value in each column. Estimates exclude top-up fees and cache writes. Speed and first-token figures are external benchmarks.

Sources and calculation details

100 requests repeats the same workload; it does not predict a growing conversation. Changing token counts updates cost only. The cost reference stays fixed when filters change.

Writing Elo assesses story generation. EQ-Bench 4 assesses emotional and social intelligence. Both are useful signals for roleplay, with different test designs. Neither measures every aspect of character chat. First-token latency is not time to a visible answer for reasoning models.

Ranking methodology