SDK v0.0.85 — Output-token limits, model scaling, and catalog updates
Turns hitting output-token limits without usable tool calls now retry up to three times instead of failing, with the default output allowance now scaling to 30% of each model's advertised limit. The model catalog has been refreshed with 49 new models, and default models have changed for 10 providers.
- A turn that hits the model's output-token limit without producing a usable tool call is now retried instead of failing the run. Reasoning or a large response could consume the whole output allowance before the agent acted, and a single
max-tokensfinish ended the run with the output-token-limit error. The runtime now retries up to three consecutive times, appending a reminder to respond concisely and split large work across tool calls; the counter resets on tool-call progress and at the start of each run, and exhaustion still fails with the existing error. Empty max-tokens responses — turns that spent the allowance with no visible output — take the same recovery path, andturn-finishedis emitted before each recovery iteration so iteration events stay paired - The default output allowance now scales with the model. When neither a request limit nor a caller default is supplied and the model advertises an output limit, the gateway uses
max(32000, floor(maxOutputTokens * 0.3))instead of a flat 32,000. The fraction can only raise the default, never lower it, so it changes nothing below roughly a 106,667-token model limit (a 128,000-token model goes from 32,000 to 38,400); the model-output and remaining-context clamps and the reasoning-budget floor still apply, and explicit request limits and caller defaults are handled as before. Longer permitted responses can raise per-turn cost and latency on large-output models - Refreshed the model catalog: still 209 providers, 6,188 to 6,237 models. The Cline Pass tier gains
cline-pass/mimo-v2.6-flashandcline-pass/mimo-v2.6-pro. The resolved default model changes for 10 providers that do not pin one inbuiltins.ts— six land on MiMo V2.6 Flash (Xiaomi and its three token plans, Empirio Labs, NanoGPT), Eden AI moves to Grok 4.7, CoreWeave to DeepSeek V4.1 Flash, CoralBricks to DeepSeek V4.1 Flash Fast, and Tempr from Claude Fable 5.1 to Gemini 3.8 Flash. If you use one of those providers without pinning a model, expect a different default
Full Changelog: sdk/sdk/v0.0.84...sdk/sdk/v0.0.85
Source: original entry ↗
More from Cline
Follow Cline to get its new changes in your feed and email digest.
Desktop v0.0.48
Added support for GPT-6 models on OpenAI and Azure OpenAI endpoints by fixing the max_tokens parameter issue. Updated default models across multiple providers and refreshed the model catalog with new CommandCode and Reflection providers.
CLI v3.0.73
GPT-6 models now work with OpenAI and Azure OpenAI endpoints after fixing a token parameter issue. Claude Sonnet 5.5 is restored as the default model when none is selected, and the model catalog has been refreshed with new providers and updated default models across multiple services.
SDK v0.0.95
OpenAI-compatible requests for GPT-6 and later models now use max_completion_tokens instead of max_tokens to prevent rejection by OpenAI and Azure. The hub daemon now tracks client launcher information as OTel resource attributes, the model catalog expanded from 6,952 to 7,041 models with new providers, and default models were updated across multiple providers with Claude Sonnet 5.5 restored as the primary recommended option.