March 25, 2026 · Cerebras
Logprobs computed before temperature scaling, so they no longer vary with temperature
Logprobs are now computed from the model's raw output, before applying temperature scaling. Previously, logprob values would vary with temperature — they now remain consistent regardless of the temperature used. This aligns with standard behavior across major inference providers and vLLM. Sampled outputs are unaffected.
Latest Cerebras Inference changes
Kimi K3.1 Pro added to Dedicated catalog, with image inputs and reasoning levels
Kimi K3, GLM 5.3 Flash and three more models added to Dedicated catalog
Service Tiers docs updated for service_tier reporting and auto and priority fallbacks
qwen-3.8-27b Developer tier rate limit raised temporarily to 750K tokens per minute
gemma-4-31b removed from Cerebras public endpoints, still on Dedicated