v1.82.1-focus-dev

LiteLLM Releases / 3/12/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical Usage

Read original →

共有:

Key Points

The v1.82.1-focus-dev release patches multiple Litellm components (completion bridge, router, proxy, streaming, and caching) to improve reliability and correctness.
It fixes handling of ResponseApplyPatchToolCall in the completion bridge by @jtsaw in PR #20913.
It updates the router to break the retry loop on non-retryable errors, contributed by @AtharvaJaiswal005 in PR #21370.
It corrects the OpenAPI schemas for the /spend/calculate and /credentials endpoints, contributed by @AtharvaJaiswal005 in PR #21369.
It preserves usage and cached_tokens in the Responses API streaming bridge and injects default_in_memory_ttl in DualCache async_set_cache and async_set_cache_pipeline, with contributions from @KeremTurgutlu and @pnookala-godaddy.

What's Changed

fix handling of ResponseApplyPatchToolCall in completion bridge by @jtsaw in #20913
fix(router): break retry loop on non-retryable errors by @AtharvaJaiswal005 in #21370
fix(proxy): fix invalid OpenAPI schema for /spend/calculate and /credentials endpoints by @AtharvaJaiswal005 in #21369
fix: preserve usage/cached_tokens in Responses API streaming bridge by @KeremTurgutlu in #22194
fix(caching): inject default_in_memory_ttl in DualCache async_set_cache and async_set_cache_pipeline by @pnookala-godaddy in #22241
fix: apply server root path to mapped passthrough route matching by @umut-polat in #22310
fix(responses): merge parallel function_call items into single assist… by @Varad2001 in #23116
fix: handle month overflow in duration_in_seconds for multi-month durations by @jnMetaCode in #23099
fix: use correct divisor when averaging TTFT in lowest-latency routing by @jnMetaCode in #23100
fix(fireworks): strip duplicate /v1 from models endpoint URL by @s-zx in #23113
fix(sagemaker): Add role assumption support for embedding endpoint by @jymmi in #20435
merge main by @Sameerlite in #23252
merge main by @Sameerlite in #23253
Litellm oss staging 03 02 2026 by @krrishdholakia in #22628
oss staging 03/09/2026 by @krrishdholakia in #23164
Litellm oss staging 02 18 2026 by @krrishdholakia in #23222
fix(vertex_ai): strip LiteLLM-internal keys from extra_body before merging to Gemini request by @Sameerlite in #23131
fix(openai): preserve reasoning_effort summary field for Responses API by @Sameerlite in #23151
fix(bedrock): populate completion_tokens_details in Responses API by @Sameerlite in #23243