Ray
2b3839d4e2
## Summary Fixes #10540. - Recognize GPT-6 Astra, Sol, and Luna as native reasoning models in `is_openai_reasoning_model()`, including Bedrock-prefixed IDs such as `us.openai.gpt-6-astra`. This prevents an explicitly configured GPT-6 `reasoning_model` from being skipped. - Add a GPT-6 Astra cookbook demonstrating streaming responses and reasoning summaries, with README instructions and a validation log. ## Type of change - [x] Bug fix - [ ] New feature - [ ] Breaking change - [ ] Improvement - [x] Model update - [ ] Other: --- ## Checklist - [x] Code complies with style guidelines - [ ] Ran format/validation scripts (`./scripts/format.sh` and `./scripts/validate.sh`) - [x] Self-review completed - [x] Documentation updated (comments, docstrings) - [x] Examples and guides: Relevant cookbook examples have been included or updated (if applicable) - [ ] Tested in clean environment - [ ] Tests added/updated (if applicable) ### Duplicate and AI-Generated PR Check - [ ] I have searched existing [open pull requests](https://github.com/agno-agi/agno/pulls) and confirmed that no other PR already addresses this issue - [ ] If a similar PR exists, I have explained below why this PR is a better approach - [ ] Check if this PR was entirely AI-generated (by Copilot, Claude Code, Cursor, etc.) --- ## Additional Notes Validation performed: - All 20 existing OpenAI reasoning checker tests passed. - Ad hoc checks confirmed 24 GPT-6 model/class combinations fail detection before the change and pass afterward. - All 96 mocked reasoning-manager executions passed across synchronous, asynchronous, streaming, and non-streaming modes. - GPT-5.6 and non-reasoning control cases remained unchanged. - The new cookbook passed Ruff lint/format, syntax, and import checks. Live Bedrock calls were not tested. The ad hoc regression checks are not included in the committed test suite. The cookbook demonstrates main-model reasoning; it does not exercise the separate `reasoning_model` stage. AI assistance was used for the cookbook, documentation, and validation.
Reasoning
Reasoning gives Agents the ability to “think” before responding and “analyze” the results of their actions (i.e. tool calls), greatly improving the Agents’ ability to solve problems that require sequential tool calls.
Reasoning Agents go through an internal chain of thought before responding, working through different ideas, validating and correcting as needed. Agno supports 3 approaches to reasoning:
- Reasoning Models
- Reasoning Tools
- Reasoning Agents and Teams
Reasoning Models
Reasoning Models are pre-trained models that are used to reason about the world. You can try any supported Agno model and if that model has reasoning capabilities, it will be used to reason about the problem.
See the examples.
Separate Reasoning Model
A powerful feature of Agno is the ability to use a separate reasoning model from the main model. This is useful when you want to use a more powerful reasoning model than the main model.
See the examples.
Reasoning Tools
By giving a model a “think” tool, we can greatly improve its reasoning capabilities by providing a dedicated space for structured thinking. This is a simple, yet effective approach to add reasoning to non-reasoning models.
See the examples.
Reasoning Agents and Teams
Reasoning Agents are a new type of multi-agent system developed by Agno that combines chain of thought reasoning with tool use.
You can enable reasoning on any Agent by providing a reasoning_model:
from agno.agent import Agent
from agno.models.openai import OpenAIChat
agent = Agent(
model=OpenAIChat(id=”gpt-4o”),
reasoning_model=OpenAIChat(id=”gpt-4o”),
)
When an Agent with a reasoning_model is given a task, the reasoning model first solves the problem using chain-of-thought. At each step, it calls tools to gather information, validate results, and iterate until it reaches a final answer. Once complete, the results are handed back to the main model to validate and provide a response.
See the examples.