Add Claude Opus 4.7 to model list

[tawnymanticore]

What does this PR do?

Adds Anthropic's new Claude Opus 4.7 to the model list with both Anthropic and OpenRouter providers. Inherits config from Opus 4.6 (json_schema structured output, vision, doc extraction, suggested for evals/data gen), with the model_id updated to claude-opus-4-7 (Anthropic) and anthropic/claude-opus-4.7 (OpenRouter).

A new thinking levels constant CLAUDE_OPUS_4_7_ANTHROPIC_THINKING_LEVELS was added to support the new xhigh and max effort levels that are exclusive to Opus 4.7 on the Anthropic API (existing CLAUDE_ANTHROPIC_EFFORT_THINKING_LEVELS is still used by older Claude models with only low/medium/high). Levels were sourced from Anthropic's effort docs, which explicitly list low, medium, high, xhigh, max for Opus 4.7. The OpenRouter provider continues to use CLAUDE_OPENROUTER_THINKING_LEVELS (which already included xhigh).

Zero-sum swap applied in a follow-up commit: featured_rank=2, editorial_notes, suggested_for_evals, and suggested_for_data_gen were removed from Opus 4.6 since Opus 4.7 now carries them. The suggested/featured count across the Claude Opus family is unchanged.

Test Results

The Anthropic suite was re-run end-to-end with ANTHROPIC_API_KEY set (the original PR couldn't run them and reported all anthropic tests as ❌ env-auth failures — those are now real PASS/FAIL results below). All five Anthropic thinking levels (low, medium, high, xhigh, max) executed against the live API and passed, validating the new CLAUDE_OPUS_4_7_ANTHROPIC_THINKING_LEVELS constant. PDF extraction and image extraction (PNG, JPEG) also passed on the Anthropic provider.

Two ⚠️ results worth noting. (1) test_data_gen_sample_all_models_providers_with_structured_output[claude_opus_4_7-anthropic] failed twice in the parallel runs with "value is not of type 'array'" — the model returned a JSON string inside the generated_samples field instead of an array. Run individually it passes immediately; this is a content-quality flake under structured_output_mode="unknown", not a config issue. (2) test_all_built_in_models_llm_as_judge[claude_opus_4_7-openrouter] fails with output_config.format.schema: For 'integer' type, properties maximum, minimum are not supported — a known Anthropic-via-OpenRouter schema constraint that also fails on Opus 4.6 (openrouter), so this is pre-existing, not a 4.7 regression.

Claude Opus 4.7 (openrouter):

• 21 passed, 14 skipped, 1 failed (pre-existing OpenRouter+Anthropic schema constraint, also fails on Opus 4.6)
• All thinking levels (none, minimal, low, medium, high, xhigh) passed

Claude Opus 4.7 (anthropic):

• 19 passed, 9 skipped (MIME types not supported — expected), 1 ⚠️ flake
• All thinking levels (low, medium, high, xhigh, max) passed
• PDF and image extraction passed

---

Claude Opus 4.7 (openrouter):
✅ test_data_gen_all_models_providers[claude_opus_4_7-openrouter]
✅ test_data_gen_sample_all_models_providers[claude_opus_4_7-openrouter]
✅ test_data_gen_sample_all_models_providers_with_structured_output[claude_opus_4_7-openrouter]
✅ test_all_built_in_models_structured_output[claude_opus_4_7-openrouter]
✅ test_all_built_in_models_structured_input[claude_opus_4_7-openrouter]
✅ test_structured_output_cot_prompt_builder[claude_opus_4_7-openrouter]
✅ test_structured_input_cot_prompt_builder[claude_opus_4_7-openrouter]
✅ test_all_models_providers_plaintext[claude_opus_4_7-openrouter]
✅ test_cot_prompt_builder[claude_opus_4_7-openrouter]
✅ test_tools_all_built_in_models[claude_opus_4_7-openrouter]
✅ test_supports_vision_is_coherent[claude_opus_4_7-openrouter]
✅ test_provider_bad_request[claude_opus_4_7-openrouter]
✅ test_extract_document_success[image/png-text_probe5-claude_opus_4_7-openrouter]
✅ test_extract_document_success[image/jpeg-text_probe6-claude_opus_4_7-openrouter]
✅ test_extract_document_success[image/jpeg-text_probe7-claude_opus_4_7-openrouter]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/none]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/minimal]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/low]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/medium]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/high]
✅ test_thinking_level_reasoning_content[ModelProviderName.openrouter/claude_opus_4_7/xhigh]
⚠️ test_all_built_in_models_llm_as_judge[claude_opus_4_7-openrouter] — pre-existing OpenRouter+Anthropic schema constraint (min/max on integer not supported); same test also fails on claude_opus_4_6-openrouter, not a 4.7 regression

Claude Opus 4.7 (anthropic):
✅ test_data_gen_all_models_providers[claude_opus_4_7-anthropic]
✅ test_data_gen_sample_all_models_providers[claude_opus_4_7-anthropic]
✅ test_all_built_in_models_llm_as_judge[claude_opus_4_7-anthropic]
✅ test_all_built_in_models_structured_output[claude_opus_4_7-anthropic]
✅ test_all_built_in_models_structured_input[claude_opus_4_7-anthropic]
✅ test_structured_output_cot_prompt_builder[claude_opus_4_7-anthropic]
✅ test_structured_input_cot_prompt_builder[claude_opus_4_7-anthropic]
✅ test_all_models_providers_plaintext[claude_opus_4_7-anthropic]
✅ test_cot_prompt_builder[claude_opus_4_7-anthropic]
✅ test_tools_all_built_in_models[claude_opus_4_7-anthropic]
✅ test_supports_vision_is_coherent[claude_opus_4_7-anthropic]
✅ test_provider_bad_request[claude_opus_4_7-anthropic]
✅ test_extract_document_success[application/pdf-text_probe0-claude_opus_4_7-anthropic]
✅ test_extract_document_success[image/png-text_probe5-claude_opus_4_7-anthropic]
✅ test_extract_document_success[image/jpeg-text_probe6-claude_opus_4_7-anthropic]
✅ test_extract_document_success[image/jpeg-text_probe7-claude_opus_4_7-anthropic]
✅ test_thinking_level_reasoning_content[ModelProviderName.anthropic/claude_opus_4_7/low]
✅ test_thinking_level_reasoning_content[ModelProviderName.anthropic/claude_opus_4_7/medium]
✅ test_thinking_level_reasoning_content[ModelProviderName.anthropic/claude_opus_4_7/high]
✅ test_thinking_level_reasoning_content[ModelProviderName.anthropic/claude_opus_4_7/xhigh]
✅ test_thinking_level_reasoning_content[ModelProviderName.anthropic/claude_opus_4_7/max]
⚠️ test_data_gen_sample_all_models_providers_with_structured_output[claude_opus_4_7-anthropic] — content flake under structured_output_mode="unknown" (model returned a stringified JSON inside generated_samples instead of the array); passes on individual re-run

Checklists

• Tests have been run locally and passed (Anthropic suite re-run with API key; OpenRouter suite passed)
• New tests have been added to any work in /lib

---

Slack thread: https://getkiln.slack.com/archives/C0AG8U78MNG/p1776354187525959

Summary by CodeRabbit

• New Features
  • Added Claude Opus 4.7 with extended thinking levels, including Extra High and Max, available via Anthropic and OpenRouter.
  • Restored Claude Opus 4.6 as a distinct option alongside 4.7.
  • OpenRouter multimodal support expanded to accept PDF (in addition to TXT, MD, JPG, PNG) with improved PDF-as-image handling.

[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: 897443a7-5c35-408d-99d7-6e9acea98606

📥 Commits

Reviewing files that changed from the base of the PR and between f968e0e and 3be434d.

📒 Files selected for processing (1)

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

🚧 Files skipped from review as they are similar to previous changes (1)

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

---

📝 Walkthrough

Walkthrough

Adds ModelName.claude_opus_4_7, a new CLAUDE_OPUS_4_7_ANTHROPIC_THINKING_LEVELS mapping (includes xhigh, max), and registers KilnModel entries for Claude Opus 4.7 and a reintroduced 4.6 with provider-specific IDs, thinking-level configs, and multimodal MIME settings in libs/core/kiln_ai/adapters/ml_model_list.py.

Changes

Cohort / File(s)	Summary
Claude Opus model registry libs/core/kiln_ai/adapters/ml_model_list.py	Added ModelName.claude_opus_4_7; introduced CLAUDE_OPUS_4_7_ANTHROPIC_THINKING_LEVELS (adds xhigh, max); updated built-in KilnModel entries to target Opus 4.7 (Anthropic/OpenRouter model_ids), added PDF to OpenRouter multimodal MIME types and multimodal_requires_pdf_as_image=True; re-added Opus 4.6 as a separate KilnModel entry with its own provider IDs and multimodal MIME list.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• Claude opus 4.6 + enabling multimodal for Opus 4.5 (+Skill update) #1008: Modifies the same model-list to add/update Claude Opus variants and provider configs.
• Qwen 3.5 and Sonnet 4.6 #1042: Adds new ModelName enum members and corresponding KilnModel entries in the same module.
• Add GLM 4.5 and 4.5 Air #493: Updates libs/core/kiln_ai/adapters/ml_model_list.py with new enum members and KilnModel entries for Claude variants.

Suggested reviewers

• sfierro
• leonardmq
• scosman
• chiang-daniel

Poem

🐰 I hopped through lines with nimble cheer,
New Opus voices now appear,
Two extra levels for thoughts to climb,
PDFs join multimodal time,
Kiln’s garden grows—now code and carrot near. 🥕✨

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically identifies the main change: adding Claude Opus 4.7 to the model list, which aligns with the PR's primary objective.
Description check	✅ Passed	The description comprehensively covers the PR objectives, implementation details, test results, and includes required checklist items, though the CLA statement is missing.
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/slack-session-8OBVl

---

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 Claude Opus 4.7 model, including its entry in the ModelName enum, a new set of thinking levels for the Anthropic provider, and the model definition for both OpenRouter and Anthropic providers. Feedback suggests ensuring the OpenRouter provider supports the full range of thinking levels, adding PDF support to the OpenRouter provider's multimodal types, and adjusting the featured_rank to avoid conflicts with older models.

[gemini-code-assist]

The openrouter provider for Claude Opus 4.7 is missing KilnMimeType.PDF in its multimodal_mime_types. Since the anthropic provider includes it (line 1959) and Claude Opus models support document extraction, it should be added here as well for consistency. Additionally, following the pattern of other recent Claude models on OpenRouter (like Sonnet 4.6 at line 1785), multimodal_requires_pdf_as_image=True should likely be set.

Suggested change

	multimodal_mime_types=[
	KilnMimeType.TXT,
	KilnMimeType.MD,
	KilnMimeType.JPG,
	KilnMimeType.PNG,
	],
	multimodal_mime_types=[
	KilnMimeType.PDF,
	KilnMimeType.TXT,
	KilnMimeType.MD,
	KilnMimeType.JPG,
	KilnMimeType.PNG,
	],
	multimodal_requires_pdf_as_image=True,

[tawnymanticore]

trying it out now

[tawnymanticore]

Addressed in 3be434d — added KilnMimeType.PDF and multimodal_requires_pdf_as_image=True to the OpenRouter provider. PDF extraction test passed on re-run (test_extract_document_success[application/pdf-text_probe0-claude_opus_4_7-openrouter]).

---

Generated by Claude Code

[coderabbitai]

🧹 Nitpick comments (1)

libs/core/kiln_ai/adapters/ml_model_list.py (1)

1919-2010: Consider extracting shared Claude Opus provider config to reduce drift risk.

Lines 1919-2010 duplicate large provider blocks between claude_opus_4_7 and claude_opus_4_6. Centralizing shared MIME lists (or shared provider defaults) would make future Claude bumps safer and smaller.

♻️ Example refactor direction

+CLAUDE_OPUS_OPENROUTER_MIME_TYPES = [
+    KilnMimeType.TXT,
+    KilnMimeType.MD,
+    KilnMimeType.JPG,
+    KilnMimeType.PNG,
+]
+
+CLAUDE_OPUS_ANTHROPIC_MIME_TYPES = [
+    KilnMimeType.PDF,
+    KilnMimeType.TXT,
+    KilnMimeType.MD,
+    KilnMimeType.JPG,
+    KilnMimeType.PNG,
+]
...
                 multimodal_mime_types=[
-                    KilnMimeType.TXT,
-                    KilnMimeType.MD,
-                    KilnMimeType.JPG,
-                    KilnMimeType.PNG,
+                    *CLAUDE_OPUS_OPENROUTER_MIME_TYPES,
                 ],
...
                 multimodal_mime_types=[
-                    KilnMimeType.PDF,
-                    KilnMimeType.TXT,
-                    KilnMimeType.MD,
-                    KilnMimeType.JPG,
-                    KilnMimeType.PNG,
+                    *CLAUDE_OPUS_ANTHROPIC_MIME_TYPES,
                 ],

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@libs/core/kiln_ai/adapters/ml_model_list.py` around lines 1919 - 2010,
Extract the repeated Claude provider configuration into shared
constants/functions to avoid drift: factor out common multimodal_mime_types
lists (e.g. a constant like CLAUDE_COMMON_MULTIMODAL_MIME_TYPES using
KilnMimeType), and extract shared KilnModelProvider defaults into helper
factories (e.g. make_claude_openrouter_provider(...) and
make_claude_anthropic_provider(...)) that accept differing fields (model_id,
available_thinking_levels, default_thinking_level, temp_top_p_exclusive,
suggested flags). Replace the duplicated KilnModelProvider blocks in KilnModel
instances (claude_opus_4_7 and claude_opus_4_6) to call these helpers and reuse
the MIME and provider-default constants (referencing
ModelName.claude_opus_4_7/4_6, ModelProviderName.openrouter/anthropic, and
CLAUDE_OPENROUTER_THINKING_LEVELS/CLAUDE_ANTHROPIC_EFFORT_THINKING_LEVELS).

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@libs/core/kiln_ai/adapters/ml_model_list.py`:
- Around line 1919-2010: Extract the repeated Claude provider configuration into
shared constants/functions to avoid drift: factor out common
multimodal_mime_types lists (e.g. a constant like
CLAUDE_COMMON_MULTIMODAL_MIME_TYPES using KilnMimeType), and extract shared
KilnModelProvider defaults into helper factories (e.g.
make_claude_openrouter_provider(...) and make_claude_anthropic_provider(...))
that accept differing fields (model_id, available_thinking_levels,
default_thinking_level, temp_top_p_exclusive, suggested flags). Replace the
duplicated KilnModelProvider blocks in KilnModel instances (claude_opus_4_7 and
claude_opus_4_6) to call these helpers and reuse the MIME and provider-default
constants (referencing ModelName.claude_opus_4_7/4_6,
ModelProviderName.openrouter/anthropic, and
CLAUDE_OPENROUTER_THINKING_LEVELS/CLAUDE_ANTHROPIC_EFFORT_THINKING_LEVELS).

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 3adfb455-0788-463c-9cc8-7e00d3fa82a1

📥 Commits

Reviewing files that changed from the base of the PR and between add1a4b and f968e0e.

📒 Files selected for processing (1)

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

[leonardmq]

Suspecting same as the bot about PDF