Skip to content

Models

The StrataCode model catalog — GLM-5.2, GLM-4.5-Air, DeepSeek V4 Flash, Qwen3.5 Plus, plus the auto router and a free lane.

Every model on the endpoint

Every model speaks the OpenAI API and is billed per token at the rate shown. Prefer stratacode/auto unless you have a reason to pin one — it routes each request to the cheapest model that can handle it, and bills you at that model’s rate.

Catalogue

  • StrataCode Auto (router) stratacode/auto
    128K ctx router

    router → GLM-5.2 / GLM-4.5-Air; billed at the model that answered

  • GLM-5.2 stratacode/glm-5.2
    128K ctx flagship

    US-hosted (DeepInfra primary, Z.ai fallback)

  • GLM-4.5-Air stratacode/glm-4.5-air
    128K ctx cheap

    Novita

  • DeepSeek V4 Flash stratacode/deepseek-v4-flash
    1M ctx cheap

    DeepSeek direct (upstream wiring: R3.1)

  • Qwen3.5 Plus stratacode/qwen3.5-plus
    128K ctx cheap

    Alibaba DashScope (upstream wiring: R3.1)

  • StrataCode Free stratacode/free
    128K ctx free

    free lane (subsidized)

Rates

  • GLM-5.2 id stratacode/glm-5.2 input $1.40 output $4.40 vs direct at list
  • GLM-4.5-Air id stratacode/glm-4.5-air input $0.20 output $1.10 vs direct at list
  • DeepSeek V4 Flash id stratacode/deepseek-v4-flash input $0.14 output $0.28 vs direct at list
  • Qwen3.5 Plus id stratacode/qwen3.5-plus input $0.40 output $2.40 vs direct at list
  • StrataCode Free id stratacode/free input $0.00 output $0.00 vs direct at list

Per million tokens, in USD. Cached input is billed at each model's published cached rate — $0.26 for GLM-5.2. These are the vendors' list prices: you never pay more here than going direct, and the router is what makes a given task cost less.

Sourcing and availability

GLM models are served through US-based inference hosts — DeepInfra primary, with failover — rather than a single origin. This is the standard, compliance-aware way to serve GLM, and it keeps a lane up if any one host degrades. Each host must pass our quality bench before it enters rotation, so a cheaper host never means quietly worse output.

We do not list models whose supply or compliance status is unsettled. Kimi is not in the catalogue. See trust and sourcing for the full policy.