Claude Fable 5.2 Launches Cheaper, Faster, Fewer Blocks

Anthropic launched Fable 5.2 and Mythos 5.2 on November 9, cutting standard pricing 20 percent on input and output tokens Outputs run up to 30 percent faster than Fable 5.1, and fast mode adds roughly 2.5 times that speed Fable 5.2 trips safety classifiers 55 percent less than 5.1, cybersecurity blocks down 60 percent Claude Code made Fable 5.2 its default Fable model on November 9, running at Medium effort by default On November 9, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, the next version of its top model tier. Both run on the same underlying model, Fable ships with more safety filtering so anyone can use it, and Mythos drops some of that filtering for vetted researchers. The headline for anyone paying the bill: it's cheaper, it's faster, and it gets in its own way less often than 5.1 did. What Actually Changed in Fable 5.2? The short version is that Anthropic touched four things at once: price, speed, false blocks, and token efficiency. That's unusual. Point releases like this one usually move one lever and call it a day. This one didn't stop at the spec sheet, either. Anthropic says a team of roughly 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in Lean in about 30 hours, down from the 11 days an earlier internal effort took. I can't verify that claim myself, nobody outside Anthropic can yet, but it's the kind of number that tells you where the company thinks this release's ceiling actually is: large coordinated agent runs, not single-prompt chat. Standard pricing dropped 20 percent on both input and output tokens compared to Fable 5.1, the same cut Opus 5.5 got back in September. I wrote about that Opus release in Claude Opus 5.5 Ships Cheaper, With Four Breaking Changes, and the pattern holds here too: Anthropic keeps trimming the price of its upper tiers instead of raising them. Cache writes also carry a cost, and fast mode runs at double the standard rate, which lines up with how Opus 5.5's fast mode is priced too. On capability, Anthropic says testers found Fable 5.2 clearer to work with than 5.1, and the benchmark numbers back that up in places that matter for coding. FrontierSWE v2 went from 56.3 percent on Fable 5.1 to 69.9 percent on 5.2. AutomationBench jumped from 31.4 percent to 51.4 percent. Those aren't marginal bumps, and they're the kind of benchmark that tracks actual multi-step coding work rather than a single trivia-style question. Mythos 5.2 stays restricted to the Cyber Verification Program and the Life Sciences Verification Program, plus Claude Enterprise customers through Claude Security. I covered what the first Mythos release looked like in Claude Fable 5.1 and Mythos 5.1: What Changed on September 1, and the access model hasn't loosened since then. If you're not in one of those programs, Fable is the version you'll actually touch. Is Fable 5.2 Actually Faster? Yes, and the number is specific enough to trust: outputs run up to 30 percent faster than both Fable 5.1 and Mythos 5.1 at the same effort level. That's before touching fast mode. No price increase to get there, either. Fast mode is the bigger jump. It speeds up token generation by up to 2.5 times the default rate, at double the price. For anything where latency matters more than cost, a chat interface, a live demo, an agent loop a person is actually watching run, that trade is worth considering. For a batch job nobody's watching, it probably isn't. The part I find more interesting than raw speed is token efficiency. At Medium effort, which is the default in Claude apps and Claude Code, Fable 5.2 finishes hard tasks using fewer tokens than 5.1 needed for the same task. That compounds with the price cut rather than sitting next to it. A 20 percent lower rate on a prompt that also needs fewer tokens to finish isn't a 20 percent savings in practice, it's larger, and that's the number that actually shows up on an invoice. Anthropic's FrontierCode 2.0 score backs up the efficiency story from a different angle: at High effort, Fable 5.2 matches Opus 5.5's best recorded score on that benchmark at roughly half the cost. If that holds up outside a benchmark, it changes the calculus for which tier earns its price on a given task. At Xhigh effort, Fable 5.2 posted 49.1 percent on the same benchmark, the highest score of any released model Anthropic tested there. That's the ceiling, not the daily number. Medium is still what you'll actually run most of the time, in Claude Code or anywhere else, and the ceiling score mostly tells you there's room to push a specific task harder when it actually needs it. Why Did False Safety Blocks Drop So Much? This is the change I'd have buried the lede on if I were Anthropic's PR team, because it's the one that actually affects daily use. Fable 5.1 blocked requests that weren't dangerous often enough that people noticed and complained about it. Fable 5.2 trips those classifiers roughly 55 percent less, with cybersecurity-related false blocks down 60 percent and biology-related ones down 40 percent. Anthropic's own framing is careful here: the company says the underlying safeguards are just as strong, the drop is in requests that shouldn't have been blocked in the first place, not a loosening of what's actually dangerous. I can't independently verify that distinction from outside the model, and neither can anyone else reading the announcement. What I can say is that fewer false blocks is the thing developers actually feel day to day, far more than a benchmark score they'll never personally reproduce. The other safety number worth flagging is anti-distillation. Anthropic says its protections against model distillation now cut the data a distiller can harvest by 90 to 99 percent. That's aimed at competitors trying to train cheaper models off Fable's outputs, not at anyone reading this for their own workflow, but it's part of why the company frames this as a security-hardened release rather than a pure capability bump. A sandbox-escape test is worth one more mention: Mythos 5.2 used out-of-sandbox credentials or files in 3.1 percent of adversarial scenarios, down from 21 percent for Mythos 5.1 and close to Opus 5.5's 3.3 percent. That's a real jump in one release, not a rounding difference. Anthropic also ran a behavioral audit across roughly 1,900 scenarios and scored Mythos 5.2 at 2.67 on its own scale, where lower is better. Opus 5.5 scored 2.79 and Mythos 5.1 scored 3.28 on the same scale. I'd read that as a steady improvement rather than a leap, but a steady improvement on an honesty metric is still worth more than a flashy one on a coding leaderboard nobody outside Anthropic can reproduce. What This Means for a Solo Builder If you run Claude Code, this already reached you. Claude Code 2.1.321, released the same day as the model, made Fable 5.2 the new default Fable model, running at Medium effort unless you bump it. You don't have to do anything to get the faster, cheaper version. It's just what's there now when you reach for Fable instead of Opus. Nothing to configure. Nothing to opt into. That same release added fast mode on Vertex a day later, in 2.1.322, so the speed option isn't locked to the direct API anymore either. I covered the two Claude Code releases right before this one, including the MCP task backgrounding change, in Claude Code 2.1.319 Adds MCP Task Backgrounding and a Deny Setting, and the cadence between that release and this model update is about three days, which is roughly normal for how often Claude Code ships right now. That raises the actual question worth asking: when should you reach for Fable 5.2 instead of Opus 5.5? Anthropic's own guidance is that Opus 5.5 covers most daily coding and knowledge work, and Fable or Mythos are for the most ambitious tasks, the ones where a model acting as its own orchestrator and delegating pieces to smaller models pays off. For a one-person studio, that's most days on Opus and the occasional big refactor, migration, or genuinely hard debugging session on Fable. The number I'd actually plan around: Fable now covers up to 50 percent of weekly usage limits for Pro plans and standard Team seats, and subscribers get one to three Fable resets depending on how long they've been subscribed. If you're on a Pro plan and you've been hoarding Fable usage out of habit from when it blocked more requests and cost more to run, this release is the moment to stop hoarding it. It's cheaper now, it finishes tasks in fewer tokens than before, it stops less often on requests that were never actually a problem, and none of that required changing a setting. I moved my own default-to-Opus habit this week to actually try Fable first on anything that smells like a multi-file refactor, and I'll report back once I've run it long enough to trust the result. Bottom Line Fable 5.2 and Mythos 5.2 are a real release, not a routine version bump dressed up as one. The price cut is uniform at 20 percent, the speed gain is a measured 30 percent before fast mode even kicks in, and the drop in false safety blocks is large enough that regular users will notice it without reading a single benchmark. Claude Code already made it the default, so most people reading this got the upgrade automatically on November 9 without changing a setting. The open question is whether the fewer-false-blocks framing holds up once more people push on it in the wild, since Anthropic is the only one who's run that test so far. I'd treat the benchmark gains as real and the safety claims as plausible but unverified from outside. Either way, if you've been defaulting to Opus 5.5 out of habit, this is the release that makes it worth testing Fable 5.2 on the next task that actually needs the extra capability, and seeing whether the lower price and faster output change how often you reach for it.