[yl231]

Adds the manual price sync with OpenRouter, and applies it once with OpenRouter's list prices of 2026-10-31. Based on #218, which needs to merge first; then this PR is retargeted to main.

How the manual path works

1. Sync prices: python scripts/pricing/sync_openrouter.py downloads OpenRouter's model list. Every profile in model_cost/model_profiles.yaml keyed by an OpenRouter id, or with an openrouter_id, takes OpenRouter's list price (never the :batch or :free tiers). The prices used are recorded in model_cost/price_snapshots/<date>.json, and model_cost.json is regenerated. --dry-run and --report print the changes without writing them.
2. Re-price the leaderboard: a separate step, done with the re-pricing tool (next PR). Syncing alone changes no score.

Pinned prices: a profile with sync: false keeps its price. That applies to models retired from OpenRouter and to models it never listed.

This run (2026-10-31)

30 prices change. These move the leaderboard most:

Model	Was	OpenRouter	Routers most affected
anthropic/claude-opus-4.7	15 / 75	5 / 25	AsiaInfo
deepseek/deepseek-v4-pro	0.435 / 0.87 (launch discount)	1.6 / 3.2	Weave, vLLM‑SR, Cross-Router
qwen/qwen3-235b-a22b-2507	0.071 / 0.10	0.09 / 0.55	Hybrid, cruq, Auto Router
deepseek/deepseek-v4-flash	0.14 / 0.28	0.091 / 0.182	several, cheaper
google/gemma-4-31b-it	0.08 / 0.35	0.2 / 0.4	KT-ModelRouter

The full list is in the commit and the snapshot file.

Pinned (sync: false):

• grok-4-1-fast-reasoning, claude-sonnet-4 and claude-3-haiku are retired from OpenRouter and kept at their last listed price.
• deepseek-reasoner is DeepSeek's API name, and the model behind it changed over time.
• agnes-2.0-flash, GLM-4-9B-0414, DeepSeek-R1-0528-Qwen3-8B and glm-4-air were never listed on OpenRouter.

Unmatched: 23 older profiles that no leaderboard router uses match no OpenRouter model. The sync lists them each run and leaves their price.

Effect on the leaderboard once re-priced: Arena score, now → re-priced at these prices.

Router	Arena
AsiaInfo	66.30 → 70.00
Weave	72.88 → 70.00
Hybrid Router	72.15 → 71.20
cruq-router	70.82 → 69.89
vLLM‑SR	74.96 → 74.24
Cross-Router	75.87 → 75.25
BARouter	74.22 → 73.67
Auto Router	70.27 → 69.80
KT-ModelRouter	76.37 → 76.08
others	within ±0.2

Not in this PR

• The daily or every-other-day automatic sync. It comes later, once this manual path is settled; this branch's earlier draft of the workflow is removed for now.
• Local-model prices.

Testing

• tests/test_price_sync.py (6) and tests/test_pricing.py (7) pass.
• pre-commit passes, including the check that model_cost.json matches the profiles.
• Re-costing every leaderboard router with the synced table gives the effect above, with no unpriced rows.

🤖 Generated with Claude Code