GPT-6 arrives claiming a double-digit lead over Fable 5.1 — treat launch-day numbers as a claim, not a result

OpenAI has presented GPT-6, positioned as a significant step over GPT-5.6 Sol and over Anthropic's Fable 5.1, with the launch claiming a benchmark advantage of more than 10% against Fable 5.1. That is the extent of what we can confirm from the material available: no breakdown of which benchmarks, which versions, which evaluation settings, or what it costs to run.

Context for why the number matters. On our own Zendoric Quality index — one third SWE-bench-Pro, one third LMArena, one third Terminal-Bench — Fable 5 has sat at 90 against GPT-5.6 Sol's 79. A verified double-digit lead over Fable 5.1 would not be an incremental step; it would flip the frontier leadership that has held for months. Claims that large deserve proportionally more scrutiny, not less.

Our reading: launch-day margins are the least reliable data point in this industry, because the benchmark set is chosen by the party making the claim. What we'll be watching are the tests that haven't saturated — SWE-bench-Pro on real repositories, Terminal-Bench on multi-step agentic work, and expert unguided cybersecurity tasks, where the spread between models is still wide enough to mean something. Cybench-style suites at pass@k near 100% tell you nothing. Neither does an Elo tuned to what voters like.

The honest position today is interest without endorsement. If GPT-6 holds this margin under independent evaluation, it's the most consequential model release of the year and a reminder that the frontier remains genuinely contested rather than settled. If it doesn't, it joins a long list of announcements whose headline number narrowed on contact with harder tasks. We'll have the measured verdict when the measurements exist.