ultralytics/actions
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update default models to Sonnet 5 and Opus 5 (#843)

Browse Source
This commit is contained in:
Glenn Jocher
2026-07-25 19:49:58 +02:00
committed by GitHub
parent 023864fc88
commit 8e85e446f8
7 changed files with 50 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
AGENTS.md
+1 -1
View File
@@ -57,7 +57,7 @@ Key flow: GitHub workflow event → `action.yml` step (gated by `github.event_na
`actions/utils/` is the shared core:
- **`github_utils.py`** — the `Action` class, the central abstraction. Initializes from GitHub Actions env vars (`GITHUB_TOKEN`, `GITHUB_EVENT_NAME`, `GITHUB_EVENT_PATH`), wraps REST (`get`/`post`/`patch`/...) and GraphQL requests with unified status checking, and provides high-level operations (PR diffs, labels, comments, discussions, alerts).
- **`openai_utils.py`** — AI provider abstraction supporting OpenAI and Anthropic. The provider/model is auto-detected from which API key env var is set; defaults live here as single source of truth (`OPENAI_MODEL_DEFAULT`, `ANTHROPIC_MODEL_DEFAULT`, `PR_REVIEW_MODEL_DEFAULT`, `MODEL_COSTS`). Also holds shared prompt-building and response sanitization.
- **`openai_utils.py`** — AI provider abstraction supporting OpenAI and Anthropic. The provider/model is auto-detected from which API key env var is set; defaults live here as single source of truth (`OPENAI_MODEL_DEFAULT`, `ANTHROPIC_MODEL_DEFAULT`, `OPENAI_REVIEW_MODEL_DEFAULT`, `ANTHROPIC_REVIEW_MODEL_DEFAULT`, `MODEL_COSTS`). Also holds shared prompt-building and response sanitization.
- **`common_utils.py`** — URL/redirect checking, diff filtering, file-skip patterns, HTML comment removal.
- **`version_utils.py`** — PyPI/pub.dev version checks used for publish gating.
README.md
+6 -6
View File
@@ -47,12 +47,12 @@ AI-powered formatting, labeling, and PR summaries for Python, JavaScript/TypeScr
Choose between [OpenAI](https://developers.openai.com/) or [Anthropic](https://www.anthropic.com/) for AI-powered features:
| Provider | Default Model | API Key |
| --------- | ------------------- | ------------------- |
| OpenAI | `gpt-5.6-luna` | `openai_api_key` |
| Anthropic | `claude-sonnet-4-6` | `anthropic_api_key` |
| Provider | Default Model | Default PR Review Model | API Key |
| --------- | ----------------- | ----------------------- | ------------------- |
| OpenAI | `gpt-5.6-luna` | `gpt-5.6-terra` | `openai_api_key` |
| Anthropic | `claude-sonnet-5` | `claude-opus-5` | `anthropic_api_key` |
The model is auto-detected based on which API key you provide. Override with the `model` input, or use `review_model` to override PR review only.
The provider is auto-detected based on which API key you provide, and PR reviews default to a stronger model than the other AI features. Override with the `model` input, or use `review_model` to override PR review only.
### 🛠️ How It Works
@@ -112,7 +112,7 @@ jobs:
openai_api_key: ${{ secrets.OPENAI_API_KEY }}
# anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
# model: gpt-5.6-luna # Optional: set model explicitly
# review_model: claude-opus-4-7 # Optional: override PR review model
# review_model: claude-opus-5 # Optional: override PR review model
brave_api_key: ${{ secrets.BRAVE_API_KEY }} # Used for broken link resolution
```
README.zh-CN.md
+6 -6
View File
@@ -47,12 +47,12 @@
可选择 [OpenAI](https://developers.openai.com/) 或 [Anthropic](https://www.anthropic.com/) 启用 AI 功能：
| 提供商 | 默认模型 | API Key |
| --------- | ------------------- | ------------------- |
| OpenAI | `gpt-5.6-luna` | `openai_api_key` |
| Anthropic | `claude-sonnet-4-6` | `anthropic_api_key` |
| 提供商 | 默认模型 | 默认 PR Review 模型 | API Key |
| --------- | ----------------- | ------------------- | ------------------- |
| OpenAI | `gpt-5.6-luna` | `gpt-5.6-terra` | `openai_api_key` |
| Anthropic | `claude-sonnet-5` | `claude-opus-5` | `anthropic_api_key` |
模型会根据提供的 API key 自动检测。可通过 `model` 输入覆盖默认模型，也可使用 `review_model` 仅覆盖 PR review 模型。
提供商会根据提供的 API key 自动检测，PR review 默认使用比其他 AI 功能更强的模型。可通过 `model` 输入覆盖默认模型，也可使用 `review_model` 仅覆盖 PR review 模型。
### 🛠️ 工作方式
@@ -112,7 +112,7 @@ jobs:
openai_api_key: ${{ secrets.OPENAI_API_KEY }}
# anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
# model: gpt-5.6-luna # Optional: set model explicitly
# review_model: claude-opus-4-7 # Optional: override PR review model
# review_model: claude-opus-5 # Optional: override PR review model
brave_api_key: ${{ secrets.BRAVE_API_KEY }} # Used for broken link resolution
```
action.yml
+2 -2
View File
@@ -76,10 +76,10 @@ inputs:
description: "Anthropic API Key"
required: false
model:
description: "AI Model - defaults to GPT-5.6 Luna or Sonnet 4.6 based on API key"
description: "AI Model - defaults to gpt-5.6-luna or claude-sonnet-5 based on API key"
required: false
review_model:
description: "AI Model for PR reviews (optional, defaults to gpt-5.6-terra with medium reasoning)"
description: "AI Model for PR reviews (optional, defaults to gpt-5.6-terra or claude-opus-5 based on API key)"
required: false
first_issue_response:
description: "Example response to a new issue"
actions/__init__.py
+1 -1
View File
@@ -28,4 +28,4 @@
# ├── test_summarize_pr.py
# └── ...
__version__ = "0.2.40"
__version__ = "0.2.41"
actions/utils/openai_utils.py
+10 -5
View File
@@ -23,8 +23,9 @@ WEB_SEARCH_CALL_COST = 0.01 # $10 per 1K calls
# Default models (single source of truth)
OPENAI_MODEL_DEFAULT = "gpt-5.6-luna"
ANTHROPIC_MODEL_DEFAULT = "claude-sonnet-4-6"
PR_REVIEW_MODEL_DEFAULT = "gpt-5.6-terra"
ANTHROPIC_MODEL_DEFAULT = "claude-sonnet-5"
OPENAI_REVIEW_MODEL_DEFAULT = "gpt-5.6-terra"
ANTHROPIC_REVIEW_MODEL_DEFAULT = "claude-opus-5"
MODEL_COSTS = { # (input, output) per 1M tokens
# OpenAI models
@@ -44,12 +45,14 @@ MODEL_COSTS = { # (input, output) per 1M tokens
# Anthropic Claude models
"claude-sonnet-4-5-20250929": (3.00, 15.00),
"claude-sonnet-4-6": (3.00, 15.00),
"claude-haiku-4-5": (1.00, 5.00),
"claude-haiku-4-5-20251001": (1.00, 5.00),
"claude-opus-4-5-20251101": (5.00, 25.00),
"claude-opus-4-6": (5.00, 25.00),
"claude-opus-4-7": (5.00, 25.00),
"claude-sonnet-5": (2.00, 10.00), # introductory pricing through 2026-08-31, then (3.00, 15.00)
"claude-opus-4-8": (5.00, 25.00),
"claude-opus-5": (5.00, 25.00),
"claude-sonnet-5": (2.00, 10.00), # introductory pricing through 2026-08-31, then (3.00, 15.00)
"claude-fable-5": (10.00, 50.00),
}
SYSTEM_PROMPT_ADDITION = """Guidance:
@@ -169,8 +172,10 @@ def _get_default_model() -> str:
def get_review_model() -> str:
"""Get model for PR reviews, using REVIEW_MODEL if set, otherwise PR_REVIEW_MODEL_DEFAULT."""
return REVIEW_MODEL or PR_REVIEW_MODEL_DEFAULT
"""Get model for PR reviews: REVIEW_MODEL if set, else the review default of the auto-detected provider."""
if REVIEW_MODEL:
return REVIEW_MODEL
return ANTHROPIC_REVIEW_MODEL_DEFAULT if _is_anthropic_model(_get_default_model()) else OPENAI_REVIEW_MODEL_DEFAULT
def _poll_openai_response(response_json: dict, headers: dict, timeout: int = 900) -> dict:
tests/test_openai_utils.py
+24 -7
View File
@@ -6,9 +6,11 @@ import pytest
import requests
from actions.utils.openai_utils import (
ANTHROPIC_MODEL_DEFAULT,
ANTHROPIC_REVIEW_MODEL_DEFAULT,
MODEL_COSTS,
OPENAI_MODEL_DEFAULT,
PR_REVIEW_MODEL_DEFAULT,
OPENAI_REVIEW_MODEL_DEFAULT,
_is_anthropic_model,
_openai_usage_cost,
_response_tool_calls,
@@ -22,9 +24,16 @@ from actions.utils.openai_utils import (
def test_default_models():
"""Test canonical default models are priced so max_cost budgets stay enforceable."""
assert OPENAI_MODEL_DEFAULT == "gpt-5.6-luna"
assert PR_REVIEW_MODEL_DEFAULT == "gpt-5.6-terra"
assert OPENAI_MODEL_DEFAULT in MODEL_COSTS # unpriced models disable max_cost budgets
assert PR_REVIEW_MODEL_DEFAULT in MODEL_COSTS
assert OPENAI_REVIEW_MODEL_DEFAULT == "gpt-5.6-terra"
assert ANTHROPIC_MODEL_DEFAULT == "claude-sonnet-5"
assert ANTHROPIC_REVIEW_MODEL_DEFAULT == "claude-opus-5"
for model in (
OPENAI_MODEL_DEFAULT,
OPENAI_REVIEW_MODEL_DEFAULT,
ANTHROPIC_MODEL_DEFAULT,
ANTHROPIC_REVIEW_MODEL_DEFAULT,
):
assert model in MODEL_COSTS # unpriced models disable max_cost budgets and misreport cost telemetry
assert MODEL_COSTS["gpt-5.6-sol"] == (5.00, 30.00)
assert MODEL_COSTS["gpt-5.6-terra"] == (2.50, 15.00)
assert MODEL_COSTS["gpt-5.6-luna"] == (1.00, 6.00)
@@ -95,9 +104,17 @@ def test_get_review_model_override():
def test_get_review_model_fallback():
"""Test review model fallback to default model."""
with patch("actions.utils.openai_utils.REVIEW_MODEL", None):
assert get_review_model() == PR_REVIEW_MODEL_DEFAULT
"""Test review model falls back to the review default of the auto-detected provider."""
with patch("actions.utils.openai_utils.REVIEW_MODEL", None), patch("actions.utils.openai_utils.MODEL", None):
with patch("actions.utils.openai_utils.ANTHROPIC_API_KEY", None):
assert get_review_model() == OPENAI_REVIEW_MODEL_DEFAULT
with patch("actions.utils.openai_utils.ANTHROPIC_API_KEY", "test-key"): # Anthropic-only setups must not
assert get_review_model() == ANTHROPIC_REVIEW_MODEL_DEFAULT # fall back to a model needing an OpenAI key
with patch("actions.utils.openai_utils.REVIEW_MODEL", None), patch(
"actions.utils.openai_utils.MODEL", "claude-sonnet-5"
):
assert get_review_model() == ANTHROPIC_REVIEW_MODEL_DEFAULT
@patch("requests.post")