LLM Gateway

Models and pricing

Every model available through the LLM Gateway, with context windows and per-token pricing.

View as Markdown

The gateway routes one model catalog across OpenAI, Anthropic, and open-weight providers. Pass any ID below as model. An unknown ID returns 400.

Prices are USD per million tokens. Cache write and cache read apply to prompt caching; models without cache-write pricing still discount cached reads.

Query the live catalog at any time:

curl https://llm.clusterbase.dev/v1/models \
  -H "Authorization: Bearer $CLUSTER_API_KEY"

Each entry includes the context window, default output limit, supported reasoning levels, supported speeds, and the standard rate card.

Current models

ModelContextInputOutputCache writeCache read
gpt-6-astra1,050,000$10.00$50.00$12.50$1.00
gpt-5.61,050,000$4.00$20.00$5.00$0.40
gpt-5.6-sol1,050,000$4.00$20.00$5.00$0.40
gpt-5.6-terra1,050,000$2.00$12.00$2.50$0.20
gpt-5.6-luna1,050,000$0.20$1.20$0.25$0.02
claude-fable-5-11,000,000$10.00$50.00$12.50$0.25
claude-fable-51,000,000$10.00$50.00$12.50$1.00
claude-opus-51,000,000$5.00$25.00$6.25$0.50
claude-sonnet-51,000,000$2.00$10.00$2.50$0.20
claude-haiku-4-5200,000$1.00$5.00$1.25$0.10
kimi-k31,048,576$3.00$15.00$0.30
qwen3p7-plus262,144$0.40$1.60$0.08
deepseek-v4-pro-08131,048,576$1.32$3.96$0.044
qwen3p8-max262,144$2.00$6.00$0.25

Aliases

Two product aliases track our recommended defaults, so you get model upgrades without a code change:

AliasCurrently routes to
chat-model-smartgpt-5.6-terra
chat-model-fastgpt-5.6-luna

Aliases run without reasoning; a request that sets reasoning_effort on an alias returns 400. Use a concrete model ID when you need reasoning control.

Fast mode

GPT-6 Astra and the GPT-5.6 family accept "speed": "fast" for lower-latency inference. Fast requests bill at twice the standard rate. All other models run at standard speed only; check supported_speeds in /v1/models.

Long-context pricing

For GPT-6 Astra and the GPT-5.6 family, a request whose input exceeds 272,000 tokens bills the whole request at multiplied rates: input at 2× and output at 1.5×.

Reasoning

Most models accept reasoning_effort (for example low, medium, high). The accepted levels differ per model — read reasoning_levels from /v1/models. claude-haiku-4-5 does not support reasoning. Reasoning output is only emitted on streaming requests.

GPT-6 Astra (gpt-6-astra) accepts low, medium, high, xhigh, and max; omitted effort defaults to medium. It does not support temperature or top_p — requests that set either are rejected with 400.

claude-fable-5-1 also accepts low, medium, high, xhigh, and max, with adaptive thinking always on; omitted effort defaults to high. It is available for Chat and the Build CLI but is not admitted for Cluster Bot (Computer) sessions.

Deprecated models

These IDs still resolve but are scheduled for removal. Migrate to a current model.

ModelInputOutputReplacement
gpt-5.5$5.00$30.00gpt-5.6
gpt-5.4$2.50$15.00gpt-5.6-terra
gpt-5.2$1.75$14.00gpt-5.6-luna
claude-opus-4-8$5.00$25.00claude-opus-5
claude-opus-4-7$5.00$25.00claude-opus-5
claude-sonnet-4-6$3.00$15.00claude-sonnet-5

/v1/models marks these with "deprecated": true.

On this page