Anthropic’s Claude Fable 5.2 answers 94.0% of the questions on SimpleQA Verified, a factual-recall test that Epoch AI runs, according to Epoch’s benchmark page. The next model, Google’s Gemini 4 Carbon, scores 77.7%, which puts Fable 5.2 16.3 points ahead by our arithmetic.
Epoch’s page, last updated on 10 November, lists 94% as the highest score on the test. Fable 5.2 also scores 18.4 points above OpenAI’s GPT-6 Astra (75.6%) and 23.2 points above Anthropic’s own Fable 5.1 (70.8%), by our arithmetic.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on how a model and the harness around it work together. It teaches the general idea only and does not discuss SimpleQA Verified.
The key insight: As we read Epoch’s leaderboard, the result stands out for its spacing: Fable 5.2 is 16.3 points clear of second place, while the next ten models sit within 8.0 points of each other, and Epoch’s review says the test also rewards a willingness to guess and is not contamination-resistant.
Epoch describes the benchmark as “1,000 factoid questions about politics, science and technology, art, sports, geography, music, and more”. Its review says the test “measures whether a model can recall obscure facts with no tool usage”, and that the questions “were filtered from OpenAI’s earlier work, SimpleQA, to be unambiguous and difficult for 2025-era models”.
On Epoch’s leaderboard, after Fable 5.2 at 94.0% (±0.8%) come Gemini 4 Carbon at 77.7%, Gemini 4 Argon at 76.0%, GPT-6 Astra at 75.6% and GPT-6.1 Sol at 73.9%. Anthropic’s Claude Opus 5.5 scores 72.2%.
The spacing is unusual. By our arithmetic, the 8.0 points between second place (77.7%) and eleventh (69.7%) are half the 16.3 points between first and second. Fable 5.2 leaves about 6.0% of questions without a correct answer, against about 22.3% for Gemini 4 Carbon, roughly 3.7x fewer.
Share of 1,000 SimpleQA Verified questions answered correctly with no tools, from Epoch AI’s leaderboard (page updated 10 November 2026, read 14 November); one run per model.
How Epoch Runs It
Epoch’s review says it runs “one turn generation, with no system prompt beyond the question and no tools”, runs each model once at a high or maximum reasoning setting, and grades answers with an AI model. It adds that, since 27 August 2026, it “appends an anti-abstention instruction”.
The review flags two limits. It says the headline number “also reflects a model’s willingness to guess, as abstentions are marked wrong”. It also says “All questions and answers have been public since 2024 as part of SimpleQA, so the benchmark is not contamination-resistant”, and that Epoch “has already flagged one likely contaminated model”.
Andrew Curran, who posted the chart on X, wrote: “There are few benchmarks like this where Fable 5.2 jaggedly jumped ahead of the entire pack”. He also wrote that he expects Anthropic to release its internal progress report next week; that is his expectation, not an announcement.
The report he refers to was announced by Anthropic co-founder Sam McCandlish on 5 November: “In the coming weeks, we plan to share a report with our own analysis of the speed of AI progress within Anthropic, including where we stand against our Responsible Scaling Policy”. We covered Anthropic’s filing language on its AI R&D threshold last week.
The Structural Read
As we read Epoch’s method, the headline score measures recall without tools on one run per model, so it is a narrow test of stored knowledge rather than of reasoning or agent work.
As we read the week’s results together, Fable 5.2 leads Epoch’s recall test while Andon Labs lists it about 42% below the leader on its business simulation, two benchmarks measuring different things.
Epoch AI, SimpleQA Verified benchmark review
“All questions and answers have been public since 2024 as part of SimpleQA, so the benchmark is not contamination-resistant”
Three Implications
FOR AI LABS Epoch lists Fable 5.2 at 94.0% against 77.7% for Gemini 4 Carbon and 75.6% for GPT-6 Astra on SimpleQA Verified.
FOR EVALUATORS Epoch says the headline number also reflects a model’s willingness to guess, as abstentions are marked wrong.
FOR READERS OF BENCHMARKS Epoch says it runs each model once with no tools, at a high or maximum reasoning setting.
The framework opens: “In any product cycle, there are two curves to follow.”
As we read it, a recall score is one point on the capability curve; the framework does not discuss SimpleQA Verified.
What Is Not Established
Epoch runs each model once, so the scores carry the stated margins (±0.8% for Fable 5.2). Epoch’s own review notes the questions have been public since 2024; the one case it flags is Qwen3-Max-Instruct, not Fable 5.2. Epoch does not say whether contamination affects Fable 5.2’s score.
Epoch does not report what share of Fable 5.2’s misses were refusals rather than wrong answers. Anthropic has not commented on the result, as far as we found, and was not contacted for this piece.
This Week In The AI Supercycle
AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source
This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.
Epoch AI lists Claude Fable 5.2 at 94.0% on SimpleQA Verified, 16.3 points above Gemini 4 Carbon by our arithmetic; Epoch’s review says the test is not contamination-resistant because its questions have been public since 2024.
A note on sourcing. Scores come from Epoch AI’s SimpleQA Verified leaderboard and benchmark review, read on 14 November 2026; Andrew Curran’s post on X flagged the chart; Sam McCandlish’s statement comes from his X post of 5 November. Point gaps and ratios are our arithmetic. Epoch and Anthropic were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
