- Add `_strip_image_b64_payloads` filter: rewrites `data[*].b64_json` in
image-gen responses to a 4-byte placeholder before the cassette is saved.
Image-edit and image-gen cassettes (193 MB / 184 MB / 104 MB / ...) will
shrink to <100 KB on next record. Tests assert response shape only, so
coverage is preserved.
- Add `_normalize_multipart_boundary` filter: replaces httpx's per-request
random multipart boundary with a fixed string in both Content-Type header
and body bytes. Audio-transcription / Whisper tests have been effectively
unmocked — every CI run hit live providers and was silently capped at
MAX_EPISODES_PER_CASSETTE=50. Both record and replay now see identical
bytes; the safe_body matcher works.
- Fix test_evals_api.py body poisoning: replace `int(time.time())` in eval
names with `hashlib.sha1(test_node_name)[:12]`, add a function-scoped
`managed_eval` fixture that creates and deletes the eval, and switch
`get_eval` / `update_eval` from `list_evals().data[0].id` (which made
the URL vary by run) to `managed_eval.id`. Net coverage gain: delete is
now actually exercised.
- Swap arxiv PDF URL in BaseOCRTest for the in-repo `dummy.pdf` (589 B)
served via sha-pinned jsdelivr.
- Swap etsystatic image URL in BaseLLMChatTest.test_image_url for the
in-repo LiteLLM logo (9.2 KB) served via the same jsdelivr pin.
- Add `tests/llm_translation/test_vcr_filters.py` with 14 unit tests
covering both new filters: replacement, idempotency, nesting, content-
length update, two-distinct-boundaries-converge-after-normalize, etc.
Cassettes recorded with the prior patterns will mismatch on the first CI
run after merge; recommend flushing the cassette Redis once (post-merge)
so re-records save under the new format from the start.
GitHub serves PDFs from raw.githubusercontent.com, github.com/.../raw/..., LFS, and Releases as application/octet-stream by deliberate anti-hotlinking policy. Anyone who passes a GitHub-hosted PDF URL as an OpenAI / Gemini / Bedrock file_id hits "unsupported MIME type 'application/octet-stream'" because _process_image_response inlines the URL with whatever Content-Type the server sent.
When the server-provided Content-Type is application/octet-stream or binary/octet-stream and the URL extension maps to a known MIME type (.pdf, .png, .jpg, etc.), trust the extension instead. Specific Content-Types (image/png, application/pdf) still win over the extension; the override only applies to generic binary types.
Also restores the Greptile SHA-pinned raw.githubusercontent.com URL on the file_id integration test so we test against the same hosting real users hit, no third-party CDN.
Resolve URL conflict: keep Greptile's commit-SHA pin for immutability, but route through jsDelivr (cdn.jsdelivr.net/gh/BerriAI/litellm@<sha>/...) so the response Content-Type is application/pdf instead of application/octet-stream. Without this, OpenAI / Gemini / Router PDF tests reject the inlined file_data with "unsupported MIME type 'application/octet-stream'".
The previous URL switch to raw.githubusercontent.com fixed Anthropic's "Unable to download" failure but caused OpenAI / Gemini / Router PDF tests to fail with "unsupported MIME type 'application/octet-stream'": those providers download the URL and inline it as data:<Content-Type>;base64,..., and raw.githubusercontent.com serves PDFs as application/octet-stream.
jsDelivr proxies the same in-repo fixture (cdn.jsdelivr.net/gh/BerriAI/litellm@main/...) and returns the correct Content-Type: application/pdf, so all providers (Anthropic forwards the URL natively; OpenAI/Gemini/Bedrock fetch and inline) get the right MIME type without changing transformer code.
The test was passing the Wikipedia URL https://upload.wikimedia.org/wikipedia/commons/2/20/Re_example.pdf as the file_id, which Anthropic's URL fetcher can no longer download (returns "Unable to download the file"). The URL is healthy for generic clients but Anthropic's fetcher fails on it deterministically, so the test has been red across PRs on litellm_internal_staging.
Switch to the in-repo fixture at tests/llm_translation/fixtures/dummy.pdf served via raw.githubusercontent.com - same fixture used elsewhere in the repo, no external CDN dependency that can block by user-agent.
* fix: use fastuuid helper across the codebase
First batch of changes, simple drop in replacement.
* second batch of changes
* fixed: script mistake on helper file
* fix(streaming_handler.py): maintain same 'created' across all chunks
Fixes https://github.com/BerriAI/litellm/issues/11437
* test: add unit test to ensure created is always the same across all chunks
* fix(types/utils.py): set a tool call id, if missing in delta tool call
Ensures stream chunk builder can reconstruct tool calls correctly
Fixes https://github.com/BerriAI/litellm/issues/11262
* fix(responses/transformation.py): support passing mcp server tool call to anthropic
allows switching between openai and anthropic for mcp tool calling
* fix(ollama/chat/transformation.py): set tool call id's when missing
* fix(anthropic/): fix regression when passing file url's to the 'file_id' parameter
add test and ensure anthropic file url's are correctly sent as 'document' blocks
* fix(vertex_and_google_ai_studio.py): Use same usage calculation function as non-streaming
Closes https://github.com/BerriAI/litellm/issues/10667
* test(test_vertex_and_google_ai_studio_gemini.py): update test
* feat(parallel_request_limiter_v2.py): add sliding window logic
allows rate limiting to work across minutes
* fix(parallel_request_limiter_v2.py): decrement usage on rate limit error
* fix(base_routing_strategy.py): fix merge from redis - preserve values in in-memory cache during gap b/w push to redis and read from redis
* fix(base_routing_strategy.py): catch the delta change during redis sync
ensures values are kept in sync
* fix(parallel_request_limiter_v2.py): update tpm tracking to use slot key logic
* fix: fix linting error
* test: update testing
* test: update tests
* test: skip on rate limit or internal server errors
* test: use pytest fixture instead
* test: bump mistral model
* fix(openai/gpt_transformation.py): handle missing filename for openai file data call
* fix(openai/gpt_transformation.py): clean handling for sync + async pdf url transformation flows
Fixes https://github.com/BerriAI/litellm/issues/10820
* build(model_prices_and_context_window.json): add 'supports_pdf_input' for all openai models which have 'vision' support
Follows openai guidelines
* feat(bedrock/chat): support cache pointing tool calls on Bedrock
Closes https://github.com/BerriAI/litellm/pull/10613
* fix: fix linting error
* add test_function_calling_with_tool_response to base llm tests
* run test suite for nova
* update test_function_calling_with_tool_response
* allowed ToolJsonSchemaBlock keys
* fix ToolJsonSchemaBlock
* add back pytest fixture
* test: test_prompt_caching
* test(base_llm_unit_tests.py): return '<thinking>' tag in response content
* fix(converse_transformation.py): extract `<thinking>` block from nova tool use response
Fixes https://github.com/BerriAI/litellm/issues/9063
* fix(factory.py): handle non-signature reasoning blocks to bedrock
pass as text input - bedrock raises ""User messages cannot contain reasoning content. Please remove the r
easoning content and try again." otherwise
* fix(main.py): Add drop params support for gpt
Fixes https://github.com/BerriAI/litellm/issues/10501
* fix(converse_transformation.py): fix linting error
* fix(utils.py): fix linting error
* test: cleanup test
* test: skip test until we have bedrock prompt caching permission
* fix(vertex_and_google_ai_studio.py): fix finish reason to be 'tool_calls' when tool call returned
Vertex returns 'Stop', openai format is 'tool calls'
* test(base_llm_unit_tests.py): bump test to assert tool calls in finish reason
* feat(fireworks_ai/chat): handle tool calling with fireworks ai correctly
Fixes https://github.com/BerriAI/litellm/issues/7209
* fix(utils.py): handle none type in message
* fix: fix model name in test
* fix(utils.py): fix validate check for openai messages
* fix: fix model returned
* fix(main.py): fix text completion routing
* test: update testing
* test: skip test - cohere having RBAC issues
* build(model_prices_and_context_window.json): add vertex ai gemini-2.5-flash pricing
* build(model_prices_and_context_window.json): add gemini reasoning token pricing
* fix(vertex_and_google_ai_studio_gemini.py): support counting thinking tokens for gemini
allows accurate cost calc
* fix(utils.py): add reasoning token cost calc to generic cost calc
ensures gemini-2.5-flash cost calculation is accurate
* build(model_prices_and_context_window.json): mark gemini-2.5-flash as 'supports_reasoning'
* feat(gemini/): support 'thinking' + 'reasoning_effort' params + new unit tests
allow controlling thinking effort for gemini-2.5-flash models
* test: update unit testing
* feat(vertex_and_google_ai_studio_gemini.py): return reasoning content if given in gemini response
* test: update model name
* fix: fix ruff check
* test(test_spend_management_endpoints.py): update tests to be less sensitive to new keys / updates to usage object
* fix(vertex_and_google_ai_studio_gemini.py): fix translation
* fix(litellm_proxy/chat/transformation.py): support 'thinking' param
Fixes https://github.com/BerriAI/litellm/issues/9380
* feat(azure/gpt_transformation.py): add azure audio model support
Closes https://github.com/BerriAI/litellm/issues/6305
* fix(utils.py): use provider_config in common functions
* fix(utils.py): add missing provider configs to get_chat_provider_config
* test: fix test
* fix: fix path
* feat(utils.py): make bedrock invoke nova config baseconfig compatible
* fix: fix linting errors
* fix(azure_ai/transformation.py): remove buggy optional param filtering for azure ai
Removes incorrect check for support tool choice when calling azure ai - prevented calling models with response_format unless on litell model cost map
* fix(amazon_cohere_transformation.py): fix bedrock invoke cohere transformation to inherit from coherechatconfig
* test: fix azure ai tool choice mapping
* fix: fix model cost map to add 'supports_tool_choice' to cohere models
* fix(get_supported_openai_params.py): check if custom llm provider in llm providers
* fix(get_supported_openai_params.py): fix llm provider in list check
* fix: fix ruff check errors
* fix: support defs when calling bedrock nova
* fix(factory.py): fix test
* test: move test to just checking async
* fix(transformation.py): handle function call with no schema
* fix(utils.py): handle pydantic base model in message tool calls
Fix https://github.com/BerriAI/litellm/issues/9321
* fix(vertex_and_google_ai_studio.py): handle tools=[]
Fixes https://github.com/BerriAI/litellm/issues/9080
* test: remove max token restriction
* test: fix basic test
* fix(get_supported_openai_params.py): fix check
* fix(converse_transformation.py): support fake streaming for meta.llama3-3-70b-instruct-v1:0
* fix: fix test
* fix: parse out empty dictionary on dbrx streaming + tool calls
* fix(handle-'strict'-param-when-calling-fireworks-ai): fireworks ai does not support 'strict' param
* fix: fix ruff check
'
* fix: handle no strict in function
* fix: revert bedrock change - handle in separate PR
* fix(internal_user_endpoints.py): cleanup unused variables on beta endpoint
no team/org split on daily user endpoint
* build(model_prices_and_context_window.json): gemini-2.0-flash supports audio input
* feat(gemini/transformation.py): support passing audio input to gemini
* test: fix test
* fix(gemini/transformation.py): support audio input as a url
enables passing google cloud bucket urls
* fix(gemini/transformation.py): support explicitly passing format of file
* fix(gemini/transformation.py): expand support for inferred file types from url
* fix(sagemaker/completion/transformation.py): fix special token error when counting sagemaker tokens
* test: fix import
* fix(anthropic/chat/transformation.py): Don't set tool choice on response_format conversion when thinking is enabled
Not allowed by Anthropic
Fixes https://github.com/BerriAI/litellm/issues/8901
* refactor: move test to base anthropic chat tests
ensures consistent behaviour across vertex/anthropic/bedrock
* fix(anthropic/chat/transformation.py): if thinking token is specified and max tokens is not - ensure max token to anthropic is higher than thinking tokens
* feat(converse_transformation.py): correctly handle thinking + response format on Bedrock Converse
Fixes https://github.com/BerriAI/litellm/issues/8901
* fix(converse_transformation.py): correctly handle adding max tokens
* test: handle service unavailable error
* fix(anthropic_claude3_transformation.py): fix amazon anthropic claude 3 tool calling transformation on invoke route
move to using anthropic config as base
* fix(utils.py): expose anthropic config via providerconfigmanager
* fix(llm_http_handler.py): support json mode on async completion calls
* fix(invoke_handler/make_call): support json mode for anthropic called via bedrock invoke
* fix(anthropic/): handle 'response_format: {"type": "text"}` + migrate amazon claude 3 invoke config to inherit from anthropic config
Prevents error when passing in 'response_format: {"type": "text"}
* test: fix test
* fix(utils.py): fix base invoke provider check
* fix(anthropic_claude3_transformation.py): don't pass 'stream' param
* fix: fix linting errors
* fix(converse_transformation.py): handle response_format type=text for converse
