Anthropic says both models generate output 30% faster than version 5.1 while costing 20% less. Here is how the numbers stack up:
- Pricing is now $8 per million input tokens and $40 per million output tokens
- Cached tokens cost 20 cents per million, compared with one dollar per million for GPT-6 Astra, which makes them 80% cheaper
- Fable 5.2 now undercuts GPT-6 Astra, its main rival
The cost cut lands in a market where customers are increasingly price-sensitive and, as the pitch implies, quick to switch models as long as performance holds up.
Fable Is Open, Mythos Stays Behind a Gate
The two models serve different audiences. Fable 5.2 targets coding, reasoning and AI agent work, and it launched to all users on Monday. Mythos 5.2 is aimed at cybersecurity and life-sciences research, and it remains in a restricted rollout open only to approved institutions in the company’s validation programs.
Anthropic is steering everyday users elsewhere. Claude Opus 5.5, released in September, remains the better fit for routine tasks, while Fable 5.2 is meant for harder problems in engineering, mathematics and robotics.
On performance, the company says the new models show improvements in areas like coding, reasoning, and vision
and communicate more clearly with users. It claims both beat competitors on coding benchmarks, and notes that more than 90% of its own code is now written by AI. An unnamed security company reportedly found that one in six confirmed bugs found by Claude Mythos 5.2 were missed by all of the other frontier models they tested.
Those are Anthropic’s figures, and no independent results accompany them.
A Self-Improvement Report Hangs Over the Launch
This is also the first release since a third-party report from the assessment group METR, which suggested the lab may be nearing RSI, the ability of AI models to carry on model training independently. METR found that internal models were speeding up Anthropic’s AI progress by roughly 2x, though it noted many uncertainties remain.
That puts a pointed question in front of investors: can Anthropic scale safely? Wall Street was initially eyeing a valuation above $2 trillion for the debut, but the target now sits between $1 trillion and $1.5 trillion as AI market sentiment shifts. These could be the company’s last big releases before it lists.
Hashlytics Take
The cached-token price is the number to watch, not the headline 20% cut. Coding and agent workloads resend the same context again and again, so a price of 20 cents against a dollar hits the bill customers actually run up. It also shows who the launch is really aimed at. A valuation target that has slipped from $2 trillion to as low as $1 trillion means Anthropic needs proof of demand before it lists, and undercutting GPT-6 Astra on price is the most legible proof it can offer. The harder sell is the safety story. Pairing a speed and cost win with a METR finding that internal models roughly double the lab’s own progress invites the obvious question of whether the slowdown its chief executive asked for last year is still the plan.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
