What Is Claude Fable 5.1? A Business Guide to Anthropic's…

<p>Anthropic released Claude Fable 5.1 in September 2026, alongside a restricted sibling model called Mythos 5.1 for vetted cybersecurity and life-sciences researchers. Both share the same underlying model with different safeguard levels. For most business purposes, Fable 5.1 is the one that matters: it's now the general-purpose flagship, and it changes the calculus on a few decisions we make with clients.</p> <h2>What it's actually good at</h2> <p>Anthropic's own benchmarks show Fable 5.1 outperforming or matching Opus 5 on most tasks while using fewer tokens to get there — 55.8–60.9% on Terminal-Bench 4.0 (agentic coding, up from 42.0% on Fable 5), 73.4% on CursorBench 3.2.0, and 60.9–65.0% on Humanity's Last Exam depending on tool access. Jane Street reported it solving more of their internal coding problems than either Fable 5 or Opus 5. Millennium had it catch a rare software crash that had eluded engineers and other models for years.</p> <h2>How it compares to GPT and Gemini</h2> <p>Independent trackers back up Anthropic's claims. Artificial Analysis' Intelligence Index puts Fable 5.1 at 66, ahead of Opus 5 (63) and GPT-5.6 Sol (61). The Vals Index has it leading at 67.87%, just ahead of Opus 5 (67.21%). Its predecessor was already dominant on coding-specific evals — SWE-bench Pro at 80.3% versus GPT-5.5's 58.6% and Gemini 3.1's 54.2% — and 5.1 extends that further. Gemini has led on pure-science benchmarks like GPQA Diamond before, so the picture isn't uniform across every category, but on coding and agentic work Fable 5.1 currently leads.</p> <h2>What BridgeBench shows</h2> <p>BridgeBench evaluates coding models across seven production-relevant categories (UI generation, security, refactoring, hallucination resistance, debugging, speed, cost efficiency) rather than one aggregate score — the more useful structure for matching a model to a specific piece of client work. As of this writing it hadn't published a Fable 5.1-specific breakdown; its most recent figures are for Fable 5 (41.5% on its reasoning suite, 47.6% on UI-bench). Worth checking their site directly for the 5.1 category breakdown before using it to decide between models on a category that matters for a given project.</p> <h2>What it costs</h2> <p>$10 per million input tokens, $50 per million output tokens, with a 75% discount on cached reads ($0.25/million). Anthropic states roughly 25% savings for typical workloads and up to 45% for agentic tasks, driven by using fewer tokens to reach the same or better result rather than a lower headline price. For context, GPT-5.5 ran roughly $8/$24 and Gemini 3.1 roughly $2/$12 per million tokens — Fable 5.1 sits at the top of the pricing range too, and the argument for it rests on the token-efficiency gap closing that on real agentic work.</p> <h2>Where this changes a client conversation</h2> <p>The old trade-off — use Opus for the hard reasoning, use something cheaper for volume — gets less clean-cut when a single model handles both well. For long-running agentic work (multi-step automations, extended debugging sessions, research tasks that run for a while unattended), matching or beating Opus while using fewer tokens is a real cost difference at agency scale, not just a benchmark footnote.</p> <h2>The honest caveat</h2> <p>Benchmark wins don't automatically transfer to your specific workflow — we've written before about why we time our own tasks instead of trusting leaderboards outright. No single model wins every category on every benchmark tracker, and BridgeBench's 5.1-specific numbers weren't available at time of writing. What's worth acting on here isn't "switch everything to Fable 5.1" but re-testing the model choice on any client pipeline that was built around the previous generation's trade-offs. The trade-offs just moved.</p>

What Is Claude Fable 5.1? A Business Guide to Anthropic's New Flagship