[yl231]

Refreshes the price table before the next leaderboard re-pricing.

DeepSeek's own prices

DeepSeek sells its models directly but not through OpenRouter, so none of its models has a DeepSeek endpoint there and the sync falls back to the hosts' median. These are now set by hand from DeepSeek's price page, with sync: false:

Model	Was (hosts' median)	Now	Note
deepseek/deepseek-v4-pro	1.6 / 3.2	1.32 / 3.96	DeepSeek's standard (peak-hour) rate; off-peak is half price
deepseek/deepseek-v4.1-flash	0.3 / 1.2	0.3 / 1.2	same, now sourced to DeepSeek
deepseek/deepseek-v3.2	0.295 / 0.88	0.28 / 0.42	DeepSeek's last V3.2 price; it no longer sells V3.2, since its API names now serve V4.1

V4 Flash: deepseek/deepseek-v4-flash stays synced. DeepSeek no longer sells it either, and the hosts' median is /bin/bash.14 / /bin/bash.28, its former DeepSeek price.

f-router's routing classifier

Qwen3-4B-Instruct-2507 runs locally at 4 bits. It's priced by the local-model rule (#266): the model isn't listed on OpenRouter, so it takes the size-curve price at 4B active parameters, ×0.8 for 4-bit weights. That gives 0.084 / 0.267, against the 0.17 / 0.25 entered with the submission.

OpenRouter sync of 2026-11-10

Model	OpenRouter id	Old	New	Basis
deepseek/deepseek-v4-flash-0731	deepseek/deepseek-v4-flash-0731	0.176 / 0.528	0.15 / 0.528	median of 19 host(s), 86% promotion undone
google/gemma-4-31b-it	google/gemma-4-31b-it	0.15 / 0.4	0.175 / 0.4	median of 8 host(s), 5% promotion undone
meta-llama/llama-3.3-70b-instruct	meta-llama/llama-3.3-70b-instruct	0.4415 / 0.715	0.59 / 0.72	median of 9 host(s)
openai/gpt-oss-120b	openai/gpt-oss-120b	0.15 / 0.6	0.145 / 0.6	median of 12 host(s)
qwen/qwen3-235b-a22b-2507	qwen/qwen3-235b-a22b-2507	0.145 / 0.775	0.145 / 0.675	median of 4 host(s)

Snapshot: model_cost/price_snapshots/2026-11-10.json. The leaderboard isn't re-priced here; that comes in the next ranking PR.

🤖 Generated with Claude Code