Wikiprompt
Blog›Benchmarks

LLM Benchmark Snapshot, September 6: Astra and Fable 5.1 Hit the Boards

The weekly photo: GPT-6 Astra debuts at 1797 Elo on webdev (+107 over the old leader), Claude Fable 5.1 takes the text arena and LiveBench, and Aider has not caught up yet.

LLM Benchmark Snapshot, September 6: Astra and Fable 5.1 Hit the Boards

One week after the densest launch window of the year, the leaderboards have spoken. This is the weekly photo for September 6, 2026: what each source says today, what changed since last week's snapshot, and where the sources disagree. Every number links to a permanent, dated page.

The headline: the launch week landed

GPT-6 Astra and Claude Fable 5.1, both released in the first days of September, are already on top of the boards:

  • Web development arena: GPT-6 Astra Max debuts at 1797 Elo, a +107 jump over last week's leader (Claude Opus 5 Max, 1691). Debut deltas this large are rare; the previous record this year was under +60. Claude Fable 5.1 Max lands second at 1762.
  • Text arena: Claude Fable 5.1 Max takes #1 with 1514, ending Opus 5 Max's reign (1505, now third behind Opus 5 High).
  • LiveBench: Fable 5.1 Max Effort edges its predecessor, 83.8 vs 83.4, with Astra Max right behind at 83.1 - three models within 0.7 points on a contamination-free suite.
  • BenchLM: Fable 5.1 first at 83.0, Astra second at 81.1. Note the odd one out: last week's BenchLM leader was Claude Mythos 5 (83.4), the restricted-access variant, which shows how access tiers complicate leaderboard reading.
  • Aider polyglot (practical code editing): unchanged - GPT-5 High still leads at 88.0%. Neither Astra nor Fable 5.1 has been benchmarked there yet; watch next week.

    • Where the sources disagree

    Arenas already crown the new pair; Aider has not tested them; LiveBench shows the top three separated by less than a point. If you read one board you get a coronation, if you read all five you get a photo finish with asterisks. That spread is the reason these snapshots exist.

    The receipts

  • Today's frozen photo: benchmark snapshot 2026-09-06
  • Last week's: snapshot 2026-08-30
  • Live boards, updated daily: text · web development · full history
  • Background on the models: GPT-6 Astra · Claude Fable 5 and Mythos 5 in the AI Wiki
    • Tags
    benchmarks·gpt-6-astra·claude-fable-5-1·snapshot·leaderboards

    Related Articles