Anthropic's Opus 5.5 Claims New State of the Art in Coding — and Costs Less
Anthropic's newest top-tier Claude model launches two months after Opus 5, beating its own larger Fable model on several benchmarks and dropping output-token prices.
Published: 2026-09-23 Category: Quick Take Sources: TechCrunch — Anthropic releases Opus 5.5
The Update
Anthropic released Opus 5.5 on Tuesday, the newest model in its most capable and expensive tier, and says it sets a new state of the art in coding and knowledge-work performance. According to the company, the model outpaces its own larger Fable model across many benchmarks and succeeded at tasks Fable failed. It also runs faster and cheaper: output tokens drop to $20 per million, down from $25 on the prior model.
Opus 5.5 also changes how the model communicates — less jargon, and important information moved to the front of messages. Anthropic says Sonnet 5.5 and Haiku 5.5 will follow "in the coming weeks" with similar gains.
The release is notable for what it isn't: Anthropic's first model launch since CEO Dario Amodei publicly embraced pacing the frontier, deliberately slowing capability advancement to let alignment catch up. Safety work on Opus 5.5 was broadly similar to predecessors, with pre-release evaluation by outside groups like METR and Frontier Design, though Anthropic flags more advanced training, monitoring, and security systems in preparation for future models.
Why This Matters
The pricing cut matters most. Opus is the premium tier, and dropping output prices while claiming better performance is a direct squeeze on the economics of high-end AI work. Combined with OpenAI halving GPT-6 Sol/Luna API prices the same morning, the two labs are clearly racing on cost-per-unit-of-capability, not just raw benchmark scores.
The Fable comparison cuts against Anthropic's own hierarchy. If a smaller, cheaper Opus 5.5 beats the larger Fable on benchmarks, it raises the question of whether the "bigger model = better" scaling assumption still holds at the top of the stack — or whether efficiency is becoming the frontier advantage.
The Tension With "Pacing"
The most interesting wrinkle is the timing. Amodei has publicly embraced slowing capability growth to match alignment progress, yet here Anthropic ships a new state-of-the-art model within two months of the previous one. That is not necessarily hypocrisy — the launch predates the full force of the pacing stance, and Anthropic frames it as the first under new infrastructure. But it illustrates the structural gap between a stated intention to pace and the competitive reality of shipping. Pacing, if it is real, will show up in the cadence and risk profile of future releases, not a single one.
Sources: TechCrunch — Anthropic releases Opus 5.5 with lower prices and Fable-level performance (Sep 22, 2026).