Add Grok 4.20 and Minimax M2.7 (Together AI)

[tawnymanticore]

What does this PR do?

Adds Grok 4.20 (OpenRouter) and Minimax M2.7 on Together AI as a new provider.

Test Results

The initial Minimax M2.7 Together AI config used json_schema, but the model ignored the schema and returned plain text (5 structured-output tests failed). Switched to the same config Minimax M2.5 uses on Together AI — json_instruction_and_object + reasoning_optional_for_structured_output=True + optional_r1_thinking parser — after which all structured-output tests pass.

One CoT test (test_structured_input_cot_prompt_builder) remains a pre-existing flake: the model collapses its reasoning and final answer into a single assistant turn, so the trace has 3 messages instead of the expected 5. This same test also fails for Minimax M2.5 on Together AI in the current main, so it is not specific to M2.7.

Minimax M2.7 (together_ai):

• 10 passed, 1 skipped, 1 pre-existing flake

---

Minimax M2.7 (together_ai):
✅ test_tools_all_built_in_models[minimax_m2_7-together_ai]
✅ test_all_built_in_models_structured_output[minimax_m2_7-together_ai]
✅ test_data_gen_sample_all_models_providers[minimax_m2_7-together_ai]
✅ test_cot_prompt_builder[minimax_m2_7-together_ai]
✅ test_all_built_in_models_structured_input[minimax_m2_7-together_ai]
✅ test_data_gen_all_models_providers[minimax_m2_7-together_ai]
✅ test_data_gen_sample_all_models_providers_with_structured_output[minimax_m2_7-together_ai]
✅ test_structured_output_cot_prompt_builder[minimax_m2_7-together_ai]
✅ test_all_built_in_models_llm_as_judge[minimax_m2_7-together_ai]
✅ test_all_models_providers_plaintext[minimax_m2_7-together_ai]
⚠️ test_structured_input_cot_prompt_builder[minimax_m2_7-together_ai] — model returns reasoning + final answer in a single turn, trace has 3 messages instead of 5. Same flake exists on Minimax M2.5 Together AI in main.

Checklists

• Tests have been run locally and passed
• New tests have been added to any work in /lib

https://claude.ai/code/session_01F1L5ryuY5t2MxQXbNVjQGj

Summary by CodeRabbit

• New Features
  • Added support for Grok 4.20 model via OpenRouter, featuring structured output, multimodal capabilities, and document extraction.
  • Expanded Minimax M2.7 with Together AI provider option, including reasoning support and data generation capabilities.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: ac8f715b-b6b0-4058-975f-0de694a2fd1e

📥 Commits

Reviewing files that changed from the base of the PR and between 9bcc35a and 37b640a.

📒 Files selected for processing (1)

• libs/core/kiln_ai/adapters/ml_model_list.py

---

📝 Walkthrough

Walkthrough

This pull request extends the model registry by adding support for the Grok 4.20 model with OpenRouter provider integration and expands Minimax M2.7 with a Together AI provider configuration, enabling additional structured output and reasoning capabilities.

Changes

Cohort / File(s)	Summary
Model Registry Extensions libs/core/kiln_ai/adapters/ml_model_list.py	Added grok_4_20 enum member to ModelName and corresponding KilnModel entry with OpenRouter provider supporting structured output, multimodal/vision capabilities, PDF handling, and document extraction. Extended Minimax M2.7 configuration with Together AI provider supporting json_instruction_and_object output, reasoning, data generation, and optional R1 thinking parser.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• Add Gemma 4, GLM 5 Turbo; add GLM-5, Kimi K2.5, MiniMax M2.5, and Qwen 3.5 on Fireworks & Together AI #1210: Adds Together AI provider for Minimax M2.7 model with json_instruction_and_object and optional reasoning configurations
• grok 4.1 fast, minimax M2.5, GLM 5 on Fireworks #1028: Adds Grok and Minimax model enum members and corresponding KilnModel entries to the model registry
• Adding support for GLM 4.7 #923: Extends ModelName enum with new model identifiers and adds corresponding built-in KilnModel configurations

Suggested reviewers

• leonardmq
• scosman

Poem

🐰 A Grok hops in, with vision so keen,
Four-point-twenty, the finest we've seen!
Together with Minimax, reasoning bright,
New models hop closer, making inference right! 🚀

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'Add Grok 4.20 and Minimax M2.7 (Together AI)' is a clear, concise description of the main changes in the pull request.
Description check	✅ Passed	The description covers what the PR does, includes test results, explains configuration decisions, and both checklist items are marked complete.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/retry-minimax-together-ai-eTIOl

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

📊 Coverage Report

Overall Coverage: 91%

Diff: origin/main...HEAD

No lines with coverage information in this diff.

---

• 📊 HTML Coverage Report - Interactive coverage report
• 📈 Diff Coverage Report - Detailed diff analysis
• Github Actions Run - View the full coverage report

[gemini-code-assist]

Code Review

This pull request adds support for the grok_4_20 model via OpenRouter and includes a new together_ai provider for the MiniMax model. A review comment suggests reordering the ModelName enum to ensure logical consistency with other model versions.

[gemini-code-assist]

The enum member grok_4_20 is inserted between grok_3_mini and grok_4_1_fast. While not strictly a bug, it is inconsistent with the alphabetical/logical ordering of the other grok models (grok_2, grok_3, grok_3_mini, grok_4_1_fast, grok_4). Consider reordering to maintain consistency.