Vals AI: DeepSeek V4.1 Flash on Fireworks is a third cheaper but 103 vs 286 tok/s

Vals AI @ValsAIOriginally posted 2026.10.24

Open original on X

Cost didn’t predict speed. The cheapest provider wasn’t the fastest, and the most expensive one wasn’t the slowest. DeepSeek V4.1 Flash on Fireworks costs a third less than DeepSeek’s API but generates at 103 tokens/sec against 286. Kimi K3 on Baseten was the priciest option but ran about five times faster than Moonshot’s API. The right provider depends on whether you need fast agents or cheap batch jobs.