Claude Opus 5.5: Fable-Level Performance at an Opus Price, and boostN Already Runs It
On 22 September 2026, Anthropic released Claude Opus 5.5 (claude-opus-5-5), the first model of the new 5.5 generation. The short version: performance on par with Claude Fable 5.1, ahead of it on several coding and agent benchmarks, at a lower price than its predecessor Opus 5.
Opus 5.5 already runs in boostN
I integrated Opus 5.5 into boostN.ai right after the release, so my customers can use it today. This article is already a first field test: Opus 5.5 wrote it, based on the sources linked below. Over the next few hours I’m testing the model on real tasks from my daily work and writing down where it holds up and where it doesn’t. I’ll share the results on LinkedIn as well.
The facts
- Price: $4 per million input tokens and $20 per million output tokens, 20% below Opus 5 ($5 / $25). Compared with Fable 5.1 ($10 / $50), that’s 60% less.
- Cache reads: $0.20 instead of $0.50 per million tokens, 60% cheaper.
- Real cost: Anthropic cites roughly 40% lower cost per typical workload than Opus 5, because the model also needs fewer tokens per task.
- Speed: per the vendor, output is more than 30% faster than Opus 5. Fast mode costs $8 / $40.
- Benchmarks (vendor figures): Terminal-Bench 4.0 at 66.4%, versus 55.8% for Fable 5.1 and 52.3% for Opus 5. Opus 5.5 also leads on CursorBench 4.0 and GDPval-AA (knowledge work).
- Context: 1 million token context window, up to 128k output tokens.
- Availability: via the Claude API, AWS, Google Cloud and Microsoft Azure, plus the Claude apps. Anthropic is also raising the five-hour limits on Pro, Max and Team.
What changes technically
Adaptive thinking, the model reasoning on its own before it answers, is always on in Opus 5.5. You now control how deep it thinks only through the effort parameter. If your API setup switched thinking off on purpose, adjust it before you migrate.
My take
The real news isn’t the price per token but the ratio: until now, Fable-level performance came at Fable prices. If the vendor numbers hold up in practice, you get that performance at a fraction of the cost. That’s exactly what I’m checking over the next few hours. The benchmarks come from Anthropic itself; independent confirmation is still pending.
According to Anthropic, Sonnet 5.5 and Haiku 5.5 will follow “in the coming weeks”.