fix(deepseek): promote reasoning_content on multi-turn default thinking models

[truecallerabreham]

TLDR

Problem this solves:

• Multi-turn DeepSeek reasoning calls fail with HTTP 400 when thinking param is omitted
• Stored reasoning_content is never promoted back onto assistant messages for default-on reasoning models

How it solves it:

• Treats default reasoning models as active in _thinking_mode_active unless explicitly disabled
• Prunes empty provider_specific_fields after promoting reasoning_content
• Preserves opt-in behavior for dual-mode models like deepseek-v3.2

User Flow

Before: a developer makes a multi-turn conversation call to deepseek-reasoner or deepseek-r1 through LiteLLM

1. Client sends POST https://litellm-domain/v1/chat/completions with model="deepseek/deepseek-reasoner" without an explicit thinking param
2. The initial assistant response returns with reasoning_content in provider_specific_fields
3. Client sends a second turn with the conversation history including the prior assistant message
4. DeepSeek API rejects the second request with HTTP 400: "The reasoning_content in the thinking mode must be passed back to the API."

After: the second request succeeds and preserves the reasoning chain

1. Client sends POST https://litellm-domain/v1/chat/completions with model="deepseek/deepseek-reasoner" without an explicit thinking param
2. The initial assistant response returns with reasoning_content in provider_specific_fields
3. Client sends a second turn with the conversation history including the prior assistant message
4. _thinking_mode_active identifies the default reasoning model and promotes reasoning_content, allowing DeepSeek API to accept the request and respond

Relevant issues

Fixes #37636

Pre-Submission checklist

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Screenshots / Proof of Fix

Before (d9a3381)

multi-turn reasoning_content promotion on default reasoning model

1. Send multi-turn request for deepseek-reasoner with prior assistant message containing provider_specific_fields={"reasoning_content": "step 1"}
2. Request transformation skips _fill_reasoning_content because _thinking_mode_active returns False when thinking param is omitted; assistant message has no reasoning_content field on wire message

After (ee2b318)

multi-turn reasoning_content promotion on default reasoning model

1. Send identical multi-turn request for deepseek-reasoner
2. _thinking_mode_active identifies default reasoning model and runs _fill_reasoning_content; assistant message has reasoning_content: "step 1" promoted and empty provider_specific_fields removed

$ pytest tests/unit/llms/deepseek/chat/test_deepseek_chat_transformation.py -v
============================= 63 passed in 0.26s ==============================

Type

🐛 Bug Fix

Caveats (if any)

Low

• Dual-mode models like deepseek-v3.2 still require thinking={"type": "enabled"} to activate reasoning mode as expected

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing truecallerabreham:fix/37636-deepseek-default-thinking-reasoning-content (446ddf8) with main (65c6495)}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge, with both previous findings resolved and no new actionable failures identified

Summary

This PR makes DeepSeek request transformation recognize default-on reasoning models through model metadata, promotes stored reasoning content during multi-turn requests, and removes empty provider-specific fields after promotion

• Adds thinking_always_on metadata for the affected native DeepSeek models
• Keeps explicit enablement and disablement behavior for dual-mode models
• Extends the existing DeepSeek transformation tests with sync, async, alias, malformed-parameter, and field-preservation coverage
• Removes the duplicate mapped test file introduced earlier

_{Reviews (3) · Last reviewed commit: "fix(deepseek): use thinking_always_on me..."}

[codecov]

Codecov Report

❌ Patch coverage is 88.46154% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/llms/deepseek/chat/transformation.py	88.46%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

[truecallerabreham]

Please review the updated commit with get_model_info metadata lookup and added tests under tests/test_litellm @greptileai

[truecallerabreham]

Added thinking_always_on to model_prices_and_context_window.json, removed hardcoded markers from transformation logic, and deleted the duplicate test file @greptileai