Anthropic releases Opus 5.5 with lower prices and Fable-level performance
Sep 22, 2026, 9:30 AM · TechCrunch
Opus 5.5 lands cheaper and faster two months after Opus 5 — and Anthropic says it beats bigger Fable on a pile of real work.
Why it matters
Anthropic shipped Claude Opus 5.5 on Tuesday, calling it state-of-the-art for coding and knowledge work in its own lineup. The company says the model outpaces the larger Fable model on many benchmarks and clears informal tasks Fable couldn’t finish — while cutting output-token price from $25 to $20 per million and running faster on less compute.
That combination matters more than another leaderboard flex. Frontier capability usually arrives with a premium. Here the top tier got cheaper and denser at the same time, two months after Opus 5 on July 24, with Sonnet 5.5 and Haiku 5.5 promised in the coming weeks.
From the desk
We’re reading this as a price-and-product story first, a pacing story second. If Anthropic’s claims hold — Fable-level or better on a lot of work, lower serving cost, snappier latency — developers get more capable agents without the usual “wait for the mid-tier refresh” delay. The communication changes (less jargon, important information up front) sound small until you live in a tool loop; verbosity is a hidden tax on every agent step.
The tension is deliberate. CEO Dario Amodei has been arguing publicly that fully addressing risk means pacing capabilities so prevention can keep up. Opus 5.5 is the first model out after that embrace of slowing the frontier. Anthropic still ships a stronger Opus. It also says biology and cybersecurity capability are comparable to Mythos, so the release carries the same class of safeguards used for Fable — limits around exploit discovery in compiled programs and recognizable biological weapons work, among other tasks. Outside evaluators including METR and Frontier Design ran pre-release checks; the company says more advanced monitoring is being built for what comes next.
Useful AI wins when the best model gets cheaper and clearer. The harm if the marketing outruns the guardrails is familiar: more people get Mythos-class cyber and bio affordances at Opus pricing before public policy capacity catches the curve Anthropic itself says takes time to build. “We expect to share more details on these efforts soon” is not a control. I’m watching independent coding and agent evals versus Anthropic’s SOTA claim, whether the $20 output price holds under load, and how the Sonnet/Haiku 5.5 wave lands — because a full-stack refresh is how this stopwatch actually moves for most users.
Context
Russell Brandom for TechCrunch, Sep 22, 2026. Opus remains Anthropic’s top, most expensive Claude tier; Sonnet is mid, Haiku is fast/cheap. Safeguard parity with Fable is Anthropic’s framing based on Mythos-comparable bio and cyber capability.
Who feels it
- Developers and agent builders
- A stronger coding/knowledge default at lower token cost — if third-party evals confirm the Fable-beating claim on their workloads.
- Enterprises on Claude
- Incentive to re-benchmark Opus for production agents sooner than a typical two-month-old top model would justify.
- Rival labs
- Pressure on price-performance narratives; “bigger” is a weaker sales line if mid-stack Opus clears Fable-shaped tasks.
- Safety and policy watchers
- First concrete release under Amodei’s pacing rhetoric — capability still advanced; safeguards and future monitoring are the open question.
What to watch
- Independent coding, agent, and knowledge-work evals versus Anthropic’s SOTA and Fable comparisons.
- Sonnet 5.5 and Haiku 5.5 releases and whether price cuts cascade down the stack.
- Detail on the “infrastructure” Anthropic says it’s building for public-policy-ready safety capacity.
- Whether Mythos/Fable-class bio and cyber safeguards hold under adversarial use at the new price point.
Companies: Anthropic