docs: clarify cost-aware model routing - #255
Conversation
📊 Benchmark Resultscoding
reasoning
chat
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5898b2dc3c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| **Use the right AI model for every task—not the most expensive one.** | ||
|
|
||
| Never send prod code to a cheap model. Never burn $20/hr on formatting. | ||
| Verdict stretches your Claude Code Max, Codex Pro, 9router, and OmniRoute setup further by routing each task to the least expensive capable model, while reserving frontier models for work that truly needs them. Configure once, use more of what you already have, and keep your best-model usage under control. |
There was a problem hiding this comment.
When users route through either documented public surface (verdict route or POST /v1/route), IntelligenceService.route() selects via select_best_model() (verdict/intelligence.py:232), whose ordering prioritizes quality, capability tier, and provider priority without considering cost (verdict/router.py:25-35). With eligible models at different prices, the selected model therefore need not be the least expensive, so this headline promise can lead users to expect cost savings that the normal routing path does not provide; qualify it as dispatcher-specific or connect the dispatcher to these routes. The declared route-flow invariant also places dispatch after ranking.
AGENTS.md reference: AGENTS.md:L40-L40
Useful? React with 👍 / 👎.
| [](LICENSE) | ||
| [](pyproject.toml) | ||
| [](LICENSE) | ||
| [](#project-status) |
There was a problem hiding this comment.
When a reader clicks the new active-development badge, #project-status has no matching heading or explicit anchor anywhere in this README, so the link does not navigate to any status information. Add the intended Project Status section or target an existing location.
Useful? React with 👍 / 👎.
| | | Typical router | **Verdict** | | ||
| |---|---|---| | ||
| | Model selection | Heuristic tiers, static allowlists | Orchestrator picks from the **OmniRoute catalog** — thousands of advertised models, liveness probe-verified in bounded samples | | ||
| | Model selection | Heuristic tiers, static allowlists | Orchestrator proposes candidates; Verdict admits only policy- and evidence-qualified options |】【。-vesm. ??? |
There was a problem hiding this comment.
The Model selection row has an accidental fourth cell containing 】【。-vesm. ???, while the table defines only three columns. GitHub-style renderers may discard the surplus cell, but other Markdown consumers or raw-source readers can expose the corrupted text, so remove it to keep the README portable and intelligible.
Useful? React with 👍 / 👎.
1 participant
Summary
Validation
git diff --checkCloses #254
🤖 Generated with Claude Code