Anthropic released Claude Opus 5.5 on September 22, 2026. It’s the first model in the new 5.5 family. According to Anthropic, it matches Claude Fable 5.1 on most work and costs 40% less to run than Opus 5.
This is mostly good news for anyone who pays for AI. For rivals trying to compete on price, it’s a problem. You now get frontier-level output from Anthropic’s mid-tier flagship, and it runs faster and costs less.
Key Intel
Anthropic reports these numbers:
- Performance. Opus 5.5 matches Claude Fable 5.1 on most tasks. Until now, Fable was the tier you paid extra for.
- Speed. Output speed is up more than 30%.
- Pricing. $4 per million input tokens and $20 per million output tokens. That’s 20% cheaper than Opus 5.
- Caching. Cache reads now cost $0.20 per million tokens, down 60%. A cache read is when the model reuses a prompt it has already processed.
- Availability. It’s live on every platform, including AWS, Google Cloud and Microsoft Azure.
- Plan limits. Five-hour usage limits went up on Pro, Max, Team and seat-based Enterprise plans. Users also get a saved rate-limit reset they can use whenever they want.
Why It Matters
The big story here is price, not smarts.
For the past year, top-tier reasoning came with top-tier prices. Teams had to decide which tasks were worth paying Fable-class rates for and which ones could make do with a weaker model. Opus 5.5 mostly removes that trade-off.
The cache cut matters more than it looks. Agent workflows send the same large system prompts, tool definitions and codebases to the model again and again. When those repeated tokens get 60% cheaper, long-running agents cost a lot less. If you run coding agents or document pipelines, check your bill next month. The savings could be real.
The 30% speed gain helps too. Faster output means shorter waits in interactive tools and quicker loops for agents that call the model dozens of times per task.
Safety Posture
Anthropic says Opus 5.5 is the strongest model it has tested so far on its automated behavioral audit, which the company calls its most comprehensive alignment test. External evaluators, including METR, tested the model before launch.
That’s worth noting because of timing. Press coverage points out this is Anthropic’s first model since CEO Dario Amodei publicly called for an AI slowdown. So the company is shipping a more capable, cheaper model while pointing to its alignment results. Expect critics to argue about whether those two things fit together.
Competitive Picture
The launch lands in a crowded week. Reports of a leaked OpenAI model, “GPT-6 Sol,” surfaced the same day. Whatever that turns out to be, Anthropic has set a price-to-performance bar others now have to meet.
Anthropic has also said Sonnet 5.5 and Haiku 5.5 are coming within weeks. If those smaller models make similar gains, the whole Claude lineup gets cheaper by the end of the year.
Action Items
Here’s what practitioners should do now:
- Re-benchmark your stack. If you send any work to Fable 5.1, test Opus 5.5 on the same tasks. You might get the same quality for much less.
- Recalculate agent costs. Long-context and heavily cached workflows get the biggest discount. Run the new numbers.
- Check latency-sensitive features. Faster output could make interactive features viable that were too slow before.
- Hold big migrations briefly. Sonnet 5.5 and Haiku 5.5 are weeks away. If you run high-volume, low-complexity work, wait and see those prices first.
- Use the new plan limits. If you’re on Pro, Max or Team, keep your saved rate-limit reset for heavy days.
Assessment
This doesn’t redefine what AI can do. It changes what it costs, and that’s what decides where AI actually gets used. When Fable-level output costs 40% less to run, a lot of projects that didn’t make financial sense suddenly do.
Watch the rollout of the rest of the 5.5 family and how competitors respond on price. Full benchmark and safety details are in Anthropic’s announcement.