v2.6.0

What's Changed

• fix(ops): forward function metadata in VersionedCall for decorator stacking by @Lyt060814 in #2858
• fix: emit OTel/Langfuse-conventional span attributes for @ops.trace tags and metadata by @mvanhorn in #2868
• fix(llm): handle empty and whitespace-only tool call arguments by @ansh-rohilla in #2887
• feat: Add MiniMax as built-in LLM provider by @octo-patch in #2859
• fix(llm): include thoughts tokens in Google streaming output tokens by @ansh-rohilla in #2888
• fix(anthropic): strip FORMAT_TOOL from re-encoded history to fix resume() HTTP 400 by @devteamaegis in #2863
• fix(llm): preserve interleaved OpenAI completion tool streams by @March-7 in #2890
• feat(providers): add RequestyProvider for Requesty's OpenAI-compatible router by @Thibaultjaigu in #2882
• test(python): add e2e tests for llm.call for multiple image formats by @samyak1729 in #2529
• fix(llm): encode ToolOutput results as JSON instead of Python repr by @ansh-rohilla in #2886
• chore: bump minor version to 2.6.0 by @willbakst in #2892

New Contributors

• @Lyt060814 made their first contribution in #2858
• @mvanhorn made their first contribution in #2868
• @ansh-rohilla made their first contribution in #2887
• @octo-patch made their first contribution in #2859
• @devteamaegis made their first contribution in #2863
• @March-7 made their first contribution in #2890
• @Thibaultjaigu made their first contribution in #2882
• @samyak1729 made their first contribution in #2529

Full Changelog: v2.5.0...v2.6.0