Skip to main content

SDK releases

What’s new

Features, fixes, and changes in each release of the Celeste Python SDK.

Latest published release

Celeste 0.11.0

March 27, 2026

See what changed
uv add celeste-ai==0.11.0
Pin this release in your project.

Published notes from GitHub · checked Sep 28, 2026

All releases
v0.11.0

Version 0.11.0

Release notes
  • Fixed

    Remove deprecated models no longer available on provider APIs (#243)

  • Fixed

    Backfill resolved_provider from model when inferred (#242)

  • New

    Add protocol= + base_url= for compatible APIs (#241)

  • Improved

    Add co-authors for tools PR (#199) (#240)

  • Fixed

    BlockingPortal thread leak in sync stream iteration (#239)

All 16 changes
  • New

    Unified tools= parameter for tool calling (#199)

  • New

    Add new model definitions for Google, OpenAI, and xAI (#213)

  • Fixed

    Warn when unsupported parameters are silently ignored (#212)

  • Fixed

    Correct model ID in README Quick Start (#211)

  • Improved

    Make _parse_content a pure parser, add Chunk type param (#203)

  • Fixed

    Enrich streaming HTTP errors with provider-specific messages (#196)

  • New

    Add extra_headers support to all client methods (#193)

  • Fixed

    Detect and raise SSE stream error events instead of silently discarding them (#144) (#192)

  • Fixed

    Empty/whitespace credentials should raise MissingCredentialsError (#190)

  • Improved

    Make ParameterMapper generic on Content type (#189)

  • Fixed

    Move _transform_output from provider _parse_content to base _predict pipeline (#187)

v0.10.0

Version 0.10.0

Release notes
  • Fixed

    OpenAI: Route streaming image edits to correct endpoint and format (#180)

  • New

    Models: Add Gemini 3.1 Pro and Claude Sonnet 4.6 (#179)

  • Improved

    Consolidate media constraints via generic base classes (#175) (#176)

  • Fixed

    Align _parse_usage return types with RawUsage contract (#174)

  • Improved

    Consolidate __init__, _build_stream_metadata, _parse_output into base Stream (#169) (#172)

All 25 changes
  • Fixed

    Catch UnicodeDecodeError in _handle_error_response (#170) (#171)

  • Improved

    Deduplicate streaming methods across all modalities (#168)

  • Improved

    Remove _build_headers() side effect in Anthropic client (#159) (#163)

  • Improved

    Add _json_headers() helper to APIMixin (#162)

  • Improved

    Add FieldMapper base class and migrate 41 simple mappers (#161)

  • Improved

    Extract _content_fields ClassVar to eliminate duplicate _build_metadata() overrides (#160)

  • Improved

    Extract Chat Completions protocol from 4 providers (#154)

  • Improved

    Migrate OpenResponses to protocol layer (#153)

  • Fixed

    Use strict=False in json.loads for structured output parsing (#152)

  • Improved

    Centralize usage/finish_reason type conversion in base classes (#151)

  • New

    Add auth passthrough to all namespace methods (#146)

  • New

    Add base64 serialization config for Output and Chunk models (#145)

  • Improved

    Fix Opus 4.6 model config for claude-code-action v1 (#141)

  • New

    Add web search support for Anthropic models (#140)

  • Improved

    Let Artifact validator handle base64 decoding across all providers (#139)

  • New

    Add Vertex AI support for all providers (#135)

  • Improved

    Add Claude Code GitHub Workflow (#136)

  • Improved

    Deps: Make google-auth optional under celeste-ai[gcp] (#133)

  • Improved

    Deps: Remove requests and ipykernel from runtime deps (#132)

  • Fixed

    Rewrite CONTRIBUTING.md with accurate workflow (#131)

v0.9.6

Version 0.9.6

Release notes
  • Fixed

    xAI: Use ImageUrl struct format for image edits (#129)

  • New

    xAI: Add Grok Imagine for image and video generation (#128)

Earlier releases 24 more
v0.9.4

Version 0.9.4

Release notes
  • Fixed

    Fix gemini-3-pro-image-preview failing with 400 error in celeste.images.edit and celeste.images.generate by using correct responseModalities (#123)

v0.9.3

Version 0.9.3

Release notes
  • New

    Images: Add Ollama images provider with flexible parameter validation (#119)

  • Improved

    Rename title field in MovieReview class (#117)

v0.9.2

Version 0.9.2

Release notes
  • Fixed

    Strip whitespace from API keys to prevent HTTP header errors (#116)

  • Improved

    Add Colab quickstart notebook (#115)

  • Improved

    Update provider logos to match implemented providers (#113)

  • Improved

    Update readme for v1 public beta launch (#112)

v0.9.0

Version 0.9.0

Release notes
  • Improved

    Test: fix integration test failures (#110)

  • Improved

    Add missing API key secrets for new providers (#109)

  • New

    V1 architecture migration - modality-centric API (#108)

v0.3.7

Version 0.3.7

Release notes
  • Fixed

    Speech generation: Return correct finish reason type in Google provider (#102)

  • Improved

    Bump version to 0.3.7 (#101)

  • Fixed

    Capabilities: Add explicit usage and finish_reason types to Output classes (#100)

v0.3.4

Version 0.3.4

Release notes
  • Improved

    Bump versions to 0.3.4 (#97)

  • New

    Add gemini-3-flash-preview model (#96)

  • New

    Core: Make capability optional in create_client (#95)

  • Fixed

    Add missing provider dependencies to capability packages (#94)

  • Improved

    Exclude pyproject.toml from integration test triggers (#92)

v0.3.3

Version 0.3.3

Release notes
  • Improved

    Bump all packages to v0.3.3

  • Fixed

    Add missing provider dependencies to capability packages

  • Improved

    Exclude pyproject.toml from integration test triggers (#92)

v0.3.2

Version 0.3.2

Release notes
  • Improved

    Bump celeste-ai, celeste-google, and speech-generation to v0.3.2 (#91)

  • New

    Google: Add Cloud TTS API and migrate speech generation (#90)

  • New

    Speech generation: Add Gradium TTS provider with WebSocket API (#89)

  • Improved

    Prepare v0.3.0 release (#88)

  • Improved

    Client: Add APIMixin abstract base for type-safe provider mixins (#87)

All 26 changes
  • Improved

    Providers: Standardize usage field mapping and add streaming mixins (#86)

  • New

    Providers: Add Cohere, ElevenLabs, and xAI API packages and migrate capabilities to mixins (#85)

  • Improved

    Rename BYTEDANCE to BYTEPLUS and migrate to provider mixins (#84)

  • New

    Providers: Add Mistral Chat API package and migrate capability to mixins (#83)

  • New

    Providers: Add OpenAI API package and migrate capabilities to mixins (#82)

  • New

    Providers: Add BFL Images API package (#81)

  • New

    Providers: Add Google Veo API package (#80)

  • New

    Providers: Add Google GenerateContent and Imagen API packages (#79)

  • Improved

    Core: StructuredOutput types, timeout increase, provider loading (#78)

  • Improved

    Text generation: Use StructuredOutput and improve API (#76)

  • Improved

    Text generation: Migrate Anthropic to mixin pattern (#75)

  • New

    Add Anthropic Messages API provider with core structured output infrastructure (#74)

  • New

    Introduce Authentication abstraction for flexible provider auth (#73)

  • Improved

    Reorganize packages into capabilities directory (#72)

  • New

    Add gpt-image-1-mini model for OpenAI image generation (#70)

  • New

    Add sora-2-2025-12-08 snapshot for OpenAI video generation (#71)

  • New

    Add seedream-4.5 model for ByteDance image generation (#69)

  • New

    Add Devstral 2 and Devstral Small 2 models (#68)

  • Improved

    Add GPT-5.2, GPT-5.2 Pro, and GPT-5.2 Instant models (#67)

  • Improved

    License: change from Apache 2.0 to MIT and enhance README (#65)

  • Improved

    Enhance README with social badges, feature highlights, and provider switching examples (#64)

v0.3.1

Version 0.3.1

Release notes
  • Improved

    Test: fix Int constraint test to accept booleans

  • Improved

    Test: improve test coverage for constraints and parameters

  • Fixed

    Gradium: Fix provider inconsistencies

  • New

    Speech generation: Add Gradium TTS provider with WebSocket API

  • New

    Speech generation: Add Gradium TTS provider (#XX)

All 28 changes
  • Improved

    Prepare v0.3.0 release (#88)

  • Improved

    Client: Add APIMixin abstract base for type-safe provider mixins (#87)

  • Improved

    Providers: Standardize usage field mapping and add streaming mixins (#86)

  • New

    Providers: Add Cohere, ElevenLabs, and xAI API packages and migrate capabilities to mixins (#85)

  • Improved

    Rename BYTEDANCE to BYTEPLUS and migrate to provider mixins (#84)

  • New

    Providers: Add Mistral Chat API package and migrate capability to mixins (#83)

  • New

    Providers: Add OpenAI API package and migrate capabilities to mixins (#82)

  • New

    Providers: Add BFL Images API package (#81)

  • New

    Providers: Add Google Veo API package (#80)

  • New

    Providers: Add Google GenerateContent and Imagen API packages (#79)

  • Improved

    Core: StructuredOutput types, timeout increase, provider loading (#78)

  • Improved

    Text generation: Use StructuredOutput and improve API (#76)

  • Improved

    Text generation: Migrate Anthropic to mixin pattern (#75)

  • New

    Add Anthropic Messages API provider with core structured output infrastructure (#74)

  • New

    Introduce Authentication abstraction for flexible provider auth (#73)

  • Improved

    Reorganize packages into capabilities directory (#72)

  • New

    Add gpt-image-1-mini model for OpenAI image generation (#70)

  • New

    Add sora-2-2025-12-08 snapshot for OpenAI video generation (#71)

  • New

    Add seedream-4.5 model for ByteDance image generation (#69)

  • New

    Add Devstral 2 and Devstral Small 2 models (#68)

  • Improved

    Add GPT-5.2, GPT-5.2 Pro, and GPT-5.2 Instant models (#67)

  • Improved

    License: change from Apache 2.0 to MIT and enhance README (#65)

  • Improved

    Enhance README with social badges, feature highlights, and provider switching examples (#64)

v0.3.0

Version 0.3.0

Release notes
  • Fixed

    Test: Use correct Google TTS model name gemini-2.5-flash-preview-tts

  • Fixed

    Ci: Rename BYTEDANCE_API_KEY to BYTEPLUS_API_KEY

  • Fixed

    Update optional dependency constraints to 0.3.0

  • Improved

    Prepare v0.3.0 release

  • Improved

    Client: Add APIMixin abstract base for type-safe provider mixins (#87)

All 26 changes
  • Improved

    Providers: Standardize usage field mapping and add streaming mixins (#86)

  • New

    Providers: Add Cohere, ElevenLabs, and xAI API packages and migrate capabilities to mixins (#85)

  • Improved

    Rename BYTEDANCE to BYTEPLUS and migrate to provider mixins (#84)

  • New

    Providers: Add Mistral Chat API package and migrate capability to mixins (#83)

  • New

    Providers: Add OpenAI API package and migrate capabilities to mixins (#82)

  • New

    Providers: Add BFL Images API package (#81)

  • New

    Providers: Add Google Veo API package (#80)

  • New

    Providers: Add Google GenerateContent and Imagen API packages (#79)

  • Improved

    Core: StructuredOutput types, timeout increase, provider loading (#78)

  • Improved

    Text generation: Use StructuredOutput and improve API (#76)

  • Improved

    Text generation: Migrate Anthropic to mixin pattern (#75)

  • New

    Add Anthropic Messages API provider with core structured output infrastructure (#74)

  • New

    Introduce Authentication abstraction for flexible provider auth (#73)

  • Improved

    Reorganize packages into capabilities directory (#72)

  • New

    Add gpt-image-1-mini model for OpenAI image generation (#70)

  • New

    Add sora-2-2025-12-08 snapshot for OpenAI video generation (#71)

  • New

    Add seedream-4.5 model for ByteDance image generation (#69)

  • New

    Add Devstral 2 and Devstral Small 2 models (#68)

  • Improved

    Add GPT-5.2, GPT-5.2 Pro, and GPT-5.2 Instant models (#67)

  • Improved

    License: change from Apache 2.0 to MIT and enhance README (#65)

  • Improved

    Enhance README with social badges, feature highlights, and provider switching examples (#64)

v0.2.12

Version 0.2.12

Release notes
  • New

    Image generation: Add BFL (Black Forest Labs) provider (#54)

  • Improved

    Remove explicit None assignments for usage token fields (#53)

  • Fixed

    Cohere usage parsing and remove deprecated models (#52)

v0.2.11

Version 0.2.11

Release notes
  • Improved

    Add XAI_API_KEY to integration test workflow (#51)

  • New

    Add XAI provider for text generation (#50)

v0.2.10

Version 0.2.10

Release notes
  • Improved

    Add Gemini 3 Pro Image Preview model support and bump versions (#49)

  • Improved

    Integrate Improved Structured Outputs in the Gemini API (#48)

v0.2.8

Version 0.2.8

Release notes
  • Improved

    Bump versions to 0.2.8 (#46)

  • Improved

    Standardize provider docstrings with capability suffixes (#44)

v0.2.7

Version 0.2.7

Release notes
  • New

    Add video-generation package with OpenAI, Google, and ByteDance providers (#43)

  • Improved

    Add Chunk type parameter to Stream class for type-safe streaming (#42)

  • Improved

    Remove redundant StrEnum type annotations from ParameterMapper (#41)

  • Improved

    Migrate Anthropic to native structured outputs API and remove raw_event metadata (#40)

  • Improved

    Bump version to 0.2.6 (#38)

All 19 changes
  • Improved

    Improve error messages and metadata handling (#39)

  • New

    Add GPT-5.1 model support (#37)

  • Improved

    Restructure finish_reason handling and add verbosity parameter (#36)

  • New

    Add ELEVENLABS provider and bump version to 0.2.5 (#35)

  • Improved

    Bump version to 0.2.4 and add speech-generation optional dependency (#34)

  • Improved

    Bump version to 0.2.2 and add image-intelligence optional dependency (#32)

  • Improved

    Improve code quality and standardize MIME types (#31)

  • Improved

    Add GPT-5 mini and nano model to OpenAI provider (#29)

  • Fixed

    Add BYTEDANCE_API_KEY to integration test workflow (#28)

  • New

    Add image-generation package with OpenAI, Google, and ByteDance providers (#27)

  • Fixed

    Preserve finish_reason in streaming final chunk (#26)

  • Improved

    Replace generic errors with custom exceptions in text-generation (#25)

  • Improved

    Implement custom exceptions for Celeste and add unit tests (#23)

  • Improved

    Standardize capability naming to kebab-case (#24)

Keep in touch

Good things,
as they happen.

New versions of Celeste, delivered to your inbox.

Release updates only. Read our privacy policy.