Sign in Start free
Catalogue

Models that do code

Trained or tuned on source. Fill-in-the-middle where the release supports it.

All chat code embeddings long-context reasoning
Model
Params
Context
ttft
tok/s
$ / M
QW Qwen2.5 72B Instruct fp8 · Qwen 72B 128K 41ms 198 $0.36 QC Qwen2.5 Coder 32B fp8 · Apache 2.0 32B 128K 27ms 318 $0.18 MS Mistral Small 3 24B fp8 · Apache 2.0 24B 32K 22ms 364 $0.14 DS DeepSeek V3 fp8 · DeepSeek 671B MoE 128K 58ms 132 $0.58 DR DeepSeek R1 Distill 32B fp8 · MIT 32B 128K 34ms 244 $0.20

Put the model next to the user.

One command, thirty-one regions, and an invoice that matches what you actually served.

Deploy a model Talk to an engineer