Claude Opus 5.5, the first release in the Claude 5.5 family, matches the performance of Claude Fable 5.1 according to Anthropic, while requiring meaningfully less compute to get there. That efficiency gain is what makes the price cut possible without Anthropic eating the difference.
The Actual Numbers
Input tokens now cost $4 per million and output tokens $20 per million, a 20% reduction from Opus 5 pricing. Cache reads dropped even further, coming in 60% cheaper than before.
- Input tokens: $4 per million (down from Opus 5 pricing)
- Output tokens: $20 per million, a 20% cut
- Cache reads: 60% cheaper
- Output generation: 30% faster
Beyond raw pricing, Anthropic says the model performs better on code migrations and one-shot deliverables, with improved alignment carried across the broader Claude portfolio.
Why Price Suddenly Matters This Much
Anthropic isn’t cutting prices in a vacuum. The company pointed directly to pressure from OpenAI and the rise of open-weight models as reasons the market has shifted. Industry observers increasingly treat price, not raw benchmark scores, as the metric that actually decides enterprise adoption.
Enterprise data requirements and growing demand for AI sovereignty are adding to that pressure. Companies aren’t just asking which model performs best anymore. They’re asking which model they can afford to run at production scale without the bill spiraling.
Open Weight Models Are Changing the Calculus
Open weight models are gaining real ground inside enterprise environments, and Nvidia’s Nemotron models specifically are described as approaching an inflection point where they become the default choice for many organizations. Rightsizing these open models, running smaller, task-specific versions instead of defaulting to the largest available, can cut inference spend substantially.
That’s the real competitive threat facing Anthropic. It’s not just closed model rivals undercutting on price. It’s the growing viability of self-hosted or open alternatives that sidestep API pricing entirely.
What Comes Next
Anthropic plans to bring similar efficiency gains to Claude Sonnet 5.5 and Claude Haiku 5.5. Opus 5.5 won’t be the cheapest frontier option on the market, but Anthropic is positioning it as a strong middle ground for teams that need frontier-level output without frontier-level pricing.
How this holds up against pricing moves from SpaceXAI and Meta’s Muse, both reportedly aiming to undercut current market rates, remains to be seen.
Hashlytics Take
A 40% cut sounds aggressive until you remember Opus 5 launched at a price point everyone already considered steep. This is Anthropic adjusting to where the market already moved, not leading it there. The more interesting number in this story isn’t the 40%, it’s the fact that open weight models like Nemotron are being described as approaching dominance inside enterprise stacks. If that holds, the real pricing pressure on Anthropic won’t come from OpenAI’s next model. It’ll come from companies realizing they don’t need a frontier API at all.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
