Skip to content

refactor(coder): use first_available model selection with multi-provider fallbacks #2996

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
dgageot wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −6
Open

refactor(coder): use first_available model selection with multi-provider fallbacks #2996

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 12 additions & 6 deletions pkg/config/builtin-agents/coder.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,16 @@
models:
default:
provider: anthropic
model: claude-opus-4-6
haiku:
provider: anthropic
model: claude-haiku-4-5
first_available:
- anthropic/claude-opus-4-8
Copy link
Copy Markdown

@docker-agent docker-agent Jun 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[MEDIUM] Speculative model IDs in default first_available list may cause runtime API failures

The new default model candidates include anthropic/claude-opus-4-8, openai/gpt-5.5, and google/gemini-3.5-flash. None of these are known public model identifiers. The first_available selector picks the first candidate whose credentials are configured — not whether the model ID is valid — so a user with Anthropic credentials will always resolve to anthropic/claude-opus-4-8, and the subsequent API call will fail if Anthropic's API does not recognize that model ID.

The old config used claude-opus-4-6 (also non-public), so this project likely uses pre-release/internal model names. If these are intentional internal identifiers, this is expected. If they are typos or speculative future names, users with cloud credentials will get runtime failures instead of a working fallback — the local dmr/ai/qwen3 fallback is only reached when no cloud credentials are configured at all.

Consider verifying these model IDs against the provider's current API catalog (e.g., Anthropic's model list, OpenAI's /v1/models, Google's Gemini API).

- openai/gpt-5.5
- google/gemini-3.5-flash
- dmr/ai/qwen3
fast:
first_available:
- anthropic/claude-haiku-4-5
- openai/gpt-5.4-mini
Copy link
Copy Markdown

@docker-agent docker-agent Jun 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[MEDIUM] Unrecognized model IDs in fast first_available list may cause runtime API failures

openai/gpt-5.4-mini and google/gemini-3.1-flash-lite are not recognized public model identifiers. The current OpenAI fast/mini model is typically gpt-4o-mini; gpt-5.4-mini is not a known offering. Similarly gemini-3.1-flash-lite does not match any known Gemini model name.

As with the default list, the first_available runtime resolves on credential availability, not model validity. A user with OpenAI credentials will select openai/gpt-5.4-mini and receive an API error at inference time if the model ID is wrong.

anthropic/claude-haiku-4-5 (unchanged from the old config) follows the same versioning pattern, so this may be an intentional pre-release naming scheme — but worth confirming that all three provider IDs are correct before shipping.

- google/gemini-3.1-flash-lite
- dmr/ai/qwen3

agents:
root:
Expand Down Expand Up @@ -185,7 +191,7 @@ agents:
agent: root

librarian:
model: haiku
model: fast
description: Web Researcher
instruction: |
You search the web for documentation, articles, and technical references.
Expand Down
Loading