Anthropic lost token volume, and the money moved down only one price tier, to OpenAI and Kimi K3. Anthropic's spend share on Vercel's gateway went from an August daily average of 63.5% to a seven-day average of 38.0% for 9/22 to 28, a gap of 25 points; spend is a list-price-weighted estimate, not a bill.
Judgment
Consensus (Vercel's 9/17 blog and others): open weights take volume, Anthropic takes the money, and any money that moves goes to the cheapest. The diary: the money moved 25 points but only one price tier down. ENTRY 121 ("big volume, little money") was downgraded on 9/14 to "the wallet has started to move".
Facts
Where it went (author's recomputation from Vercel's official export, August daily average to the 9/22 to 28 seven-day average): Moonshot +10.2, OpenAI +9.1, DeepSeek +3.1, Google +2.7 points. Anthropic's token share fell from 22% to 7.7%. Opus 5.5 (launched 9/22; the company says typical workloads save 40%) explains at most 4 points (self-computed). The export gives shares only, so buyers leaving cannot be told apart from dilution by new workloads.
What to do
For investors: do not read gateway share as Anthropic revenue; among Chinese models, do not back the cheapest, back the ones that can collect above $10/M on output. For founders: move agent workloads first, and look at Harvey's post-training on an open base model to repair gross margin (the margin turnaround is Bloomberg citing people familiar; the company declined to comment).
Falsifier (verbatim from the diary footnote)
Two tests, ruled separately. The ruling source for both is Vercel's official labs export recomputed with the repository script scripts/vercel-lab-share.py, with readings published; whichever triggers first decides; a 30-day window may not end before the start date, and 9/29 is counted at its complete value from a later re-pull. (1) "Volume was lost, the wallet is turning": starting 2026-09-29, through 2026-12-31, if Anthropic's average spend share over any consecutive 30-day window is not lower than 53% (it was 46.4% for 8/30 to 9/28; the ✗ line and the verified line sit about 6.5 points above and below the current value), mark ✗ (a pulse, not a trend; token share is published alongside; written back to ENTRY 121's parent thesis); if the average over any consecutive 30-day window is not higher than 40%, mark verified; if neither triggers by expiry, mark unresolved and close this test; until then mark ⏳ and publish readings.
Disclosure: This diary is written and verified with Claude; discount any judgment about Anthropic, favourable or not; The author is a Vercel customer; Vercel's leaderboard 'spend' is a list-price-weighted estimate, not billing; FutureX Capital has fully exited Meituan (confirmed by the author on Oct 10, 2026), so it no longer has an indirect interest in Moonshot AI through Meituan.