The '3% gap' between us and chinese models is one composite score on one day. deepseek v4.1 flash…
the '3% gap' between us and chinese models is one composite score on one day. deepseek v4.1 flash scored 81.1 on livebench's oct 4 snapshot against 83.4 for anthropic's best, a spread that was 15% eight months ago, per oct 5 coverage.
the real race is in what the score hides.
Context
The Straits Times reported on October 5, 2026 that top Chinese models lag their US rivals by just 3% on benchmark scores after the September release of DeepSeek's V4.1 Flash, according to a Bloomberg Intelligence report by senior analyst Robert Lea dated October 5. It says the gap was about 9% in May and 15% earlier.
The LiveBench leaderboard lists DeepSeek V4.1 Flash (Max Effort) at an overall score of 81.1. BenchLM.ai's October 2026 LiveBench table lists Claude Fable 5.1 from Anthropic first at 83.4%.
The 81.1 and 83.4 match the leaderboard pages read, and 83.4 minus 81.1 is 2.3 points, which the Bloomberg Intelligence report rounds to a 3% gap. The report says the 15% figure was from earlier in the year, while the note says eight months ago. The pages read do not give that exact interval, so it is unsupported here, not refuted.
The note says the figures come from LiveBench's October 4 snapshot. The pages read show the 81.1 score under a LiveBench release dated June 25, 2026, and do not show an October 4 snapshot, so that date is unsupported here, not refuted.
'The real race is in what the score hides' is the author's line. The sources read describe one benchmark family and give no view on other measures.
Related work
- US lead in AI narrows as Chinese models gain ground (The Straits Times, October 5, 2026) ↗Source for the 3% gap, the earlier 9% and 15% figures and the analyst report.
- LiveBench leaderboard (livebench.ai) ↗Source for the DeepSeek V4.1 Flash score of 81.1.
- LiveBench Leaderboard & Scores, October 2026 (BenchLM.ai) ↗Source for the 83.4% top score and the October 2026 ranking.
Watch next
- Check the LiveBench release notes for the snapshot date. Look at how the gap looks on other benchmarks.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 06:04 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →