Fireworks AI Batch
Batch inference billed at 50% of serverless on both input and output across the open catalog
Fireworks AI Batch scores 6.5/10 for free value in LLM. The main constraint: Paid — 50% of serverless rates. Batch inference billed at 50% of serverless on both input and output across the open catalog.
Free value score
6.5 / 10Show the math
Key metrics
What you get
Paid — 50% of serverless rates. Batch inference billed at 50% of serverless on both input and output across the open catalog
Halves all serverless per-token costs (e.g. ~2M Kimi K2.6 input/$1 batch)
Gotchas
No gotchas documented yet. Contribute one if you know a constraint we missed.
Sources
-
"Batch inference is billed at 50% of serverless pricing on both input and output."
https://docs.fireworks.ai/serverless/pricing -
"Kimi K2 Thinking achieves a 67 in the Artificial Analysis Intelligence Index"
https://x.com/ArtificialAnlys/status/1986911675820446013
Alternatives
Same category · similar quotaDirt-cheap open-model PAYG with disclosed quants: GLM-5.2 (Q8_0) $0.50/$2.20 per 1M, Kimi-K2.7-code (int4) $0.55/$2.25, GLM-5.1 $0.45/$2.15, Qwen3.5-397B (Q4_0) $0.35/$1.75, DeepSeek-V4-Pro $0.35/$0.80; OpenAI-compatible endpoint
Cheapest GLM 5.2 API verified (Jul 2026): $0.43/M input, $1.35/M output, $0.08/M cache read, 1M context; Kimi K2.7 Code $0.43/$1.79 (+HighSpeed tier); OpenAI- AND Anthropic-compatible gateway, no rate/concurrency limits, production allowed