Skip to content

release: 1.21.1 — re-sync Router Core to 3aca0e1 (Gemini 4 Argon capabilities) - #92

Closed
VickyXAI wants to merge 2 commits into
mainfrom
chore/router-core-3aca0e1
Closed

VickyXAI wants to merge 2 commits into
mainfrom
chore/router-core-3aca0e1

Conversation

@VickyXAI

@VickyXAI VickyXAI commented Oct 20, 2026 •

Copy link
Copy Markdown
Contributor

Ports router-core 2cd5a10 → 3aca0e1 (router-core #9) into blockrun_llm/router_core/. This is a hand port: the repo has no sync script, so I followed the #87 pattern.

  • model_capabilities.py adds the google/gemini-4-argon entry with supports_tools: False, so it fails closed for tool requests. It sits in upstream's order with upstream's comment.
  • The upstream commit markers in __init__.py, config.py and the two parity-test docstrings move to 3aca0e1.
  • tests/unit/router_core_decisions.snapshot.json is copied again from upstream 3aca0e1, and the result is byte-identical: no diff.

Routing unchanged:

Version 1.21.1, set in VERSION, pyproject.toml and __version__. The changelog also covers the README line from #89.

Tests: pytest tests/unit passes 1,342.

1bcMax added 2 commits October 20, 2026 14:12
…rgon capabilities)

Port router-core 2cd5a10..3aca0e1 (#9): model_capabilities.py gains
google/gemini-4-argon (1M context, 65,536 output, vision, supports_tools
False -> fails closed). Upstream config.ts and every vitest file are
unchanged in that range; the decision fixture is re-copied and byte-identical
(88/88 reproduce). Commit markers move to 3aca0e1.
@VickyXAI

Copy link
Copy Markdown
Contributor Author

Superseded by #90, which already pins router-core 8773a99 (a descendant of 3aca0e1, so the Gemini 4 Argon capability entry is shipped). Merging this would roll the pin back and undo router-core #10.

@VickyXAI VickyXAI closed this Oct 20, 2026
@VickyXAI
VickyXAI deleted the chore/router-core-3aca0e1 branch October 20, 2026 19:18
VickyXAI added a commit that referenced this pull request Oct 27, 2026
…t a skip (#96)

Port of @blockrun/llm #92 (review of the TS #89 port): settlement_skipped()
ignores a marker when the response proves the payment settled (PAYMENT-RESPONSE
not failed with a transaction or a positive amount / extra.chargedAmount, or
X-Payment-Receipt), writing the conflict to stderr; negative values are not a
marker. blockrun.ai's free-fallback receipt ({success: true, transaction: ""},
no amount) stays skipped. 8 new helper tests fail on main.

Co-authored-by: 1bcMax <viewitter@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant