Anthropic shipped Claude Opus 5.5 on September 22, 2026, and it's a strange kind of launch: the headline isn't a new capability ceiling, it's a new price floor. Opus 5.5 is pitched as performing at roughly the level of Claude Fable 5.1 — Anthropic's actual top-tier, Mythos-class model — on most everyday work, while costing 40% less to run than the outgoing Opus 5 and running over 30% faster.
That's worth pausing on, because it's an unusual move even by Anthropic's own recent playbook. In July 2026, Opus 5 was sold as near-Fable performance at roughly half of Fable's price. Opus 5.5 pushes that further: Fable-level performance at about 40% of Fable 5.1's list price, with cheaper cached tokens on top. It's the first release in what Anthropic says will become a full 5.5 family, with Sonnet 5.5 and Haiku 5.5 expected within weeks.
Quick summary: Claude Opus 5.5 launched September 22, 2026 at $4 per million input tokens and $20 per million output tokens — 20% below Opus 5's $5/$25 — with cache reads cut 60% to $0.20 per million tokens. Anthropic says it matches Fable 5.1 on most work and beats it on agentic coding, knowledge work, computer use, chart recognition, and multidisciplinary reasoning. Output speed is up over 30% versus Opus 5. It's live on the Anthropic API, AWS, Google Cloud, and Microsoft Azure; Sonnet 5.5 and Haiku 5.5 are expected in the coming weeks.
What Opus 5.5 actually is
To place this correctly you need Anthropic's current model hierarchy. As of mid-2026, Anthropic sells two tiers above the classic Opus/Sonnet/Haiku ladder: Fable, which is generally available, and Mythos, which is gated behind trusted-access programs. Fable and Mythos reportedly share the same underlying model; Fable simply routes flagged cybersecurity, biology/chemistry, and distillation-adjacent requests to a less capable fallback. Fable 5.1 is, in other words, Anthropic's real frontier model right now — not Opus. Sources: Claude Fable 5 and Claude Mythos 5 — Anthropic, Anthropic Releases Claude Fable 5 and Claude Mythos 5 — MarkTechPost.
So when Anthropic says Opus 5.5 performs at 'roughly Fable 5.1's level on most work,' it's saying a second-tier model has caught up to last quarter's flagship on typical tasks, at 40% of Fable's list price. That's the real story: Opus 5.5 isn't trying to be a new ceiling, it's trying to make Fable 5.1 unnecessary for most day-to-day agentic coding, knowledge work, and computer-use tasks — while Fable and the gated Mythos tier stay reserved for the work that genuinely needs the extra headroom.
Pricing, in full
Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, a 20% cut from Opus 5's $5/$25. Cache reads fall 60%, from Opus 5's rate down to $0.20 per million tokens — a meaningful change for anyone running long, multi-turn agentic sessions where cached context dominates the bill. Anthropic frames the combined effect as roughly a 40% reduction in cost for typical workloads. For comparison, Fable 5.1 is priced at $10 input / $50 output per million tokens, with cache reads at $0.25 per million — meaning Opus 5.5 runs at roughly 40% of Fable 5.1's per-token list price, or about 60% cheaper depending on the mix of cached versus fresh tokens.
| Pricing | Opus 5 | Opus 5.5 | Fable 5.1 |
|---|---|---|---|
| Input / 1M tokens | $5 | $4 | $10 |
| Output / 1M tokens | $25 | $20 | $50 |
| Cache reads / 1M tokens | not published | $0.20 | $0.25 |
| Output speed vs Opus 5 | baseline | 30%+ faster | n/a |
The benchmark numbers — and the independent check
On the benchmarks Anthropic published itself, Opus 5.5 beats Fable 5.1 outright on several fronts: FrontierCode v1.1 Main (54.4% vs. 50.3%) and CursorBench 4.0 (57.8% vs. 51.8%) are the two clearest, apples-to-apples wins Anthropic reported. On Terminal-Bench 4.0, Anthropic put Opus 5.5 at 66.4%, which it says is roughly ten points clear of both Fable 5.1 and GPT-6 Astra. Anthropic also claims wins on knowledge work, computer use, and visual chart recognition, though it hasn't published every underlying score.
That's Anthropic grading its own homework, and the independent numbers tell a slightly more modest story. Coverage citing Artificial Analysis's own Terminal-Bench 4.0 run put Opus 5.5 at 59.6% — well below Anthropic's 66.4% figure, and reportedly roughly level with GPT-6 Astra rather than ten points ahead of it. On Artificial Analysis's own Intelligence Index, Opus 5.5 reportedly scored 58 against Fable 5.1's 53, which does support the 'catching up to Fable' claim in broad terms, even if the margin on any single agentic benchmark is closer than Anthropic's own charts suggest.
| Benchmark | Opus 5.5 | Fable 5.1 | Basis |
|---|---|---|---|
| FrontierCode v1.1 Main | 54.4% | 50.3% | Anthropic |
| CursorBench 4.0 | 57.8% | 51.8% | Anthropic |
| Terminal-Bench 4.0 | 66.4% | ~10 pts lower (exact score not published) | Anthropic |
| Terminal-Bench 4.0 (independent) | 59.6%, reportedly level with GPT-6 Astra | not published | Artificial Analysis, per reports |
| Intelligence Index (independent) | 58 | 53 | Artificial Analysis, per reports |
One more detail worth flagging for anyone who's used Claude models for writing: multiple outlets reported Anthropic also said Opus 5.5 reduces the stiff, repetitive-tic 'Claudish' writing style people have complained about since earlier Claude generations — a qualitative claim, not a benchmarked one, so treat it as a direction rather than a solved problem until you've tested it on your own prompts. Sources: Claude Opus 5.5 matches Fable 5.1 performance — the-decoder.
New flagship, or a cheaper Fable stand-in?
So which is it? Based on the positioning Anthropic has taken and how the pricing tiers stack up, Opus 5.5 reads as the second, not the first: it's not dethroning Fable 5.1 as Anthropic's most capable public model, and it isn't marketed that way. It's a cost/speed optimization of the Opus line that happens to close most of the gap to last quarter's flagship on the workloads most people actually run — coding, knowledge work, computer use — while Fable 5.1 (and the gated Mythos 5.1) remain the models Anthropic points to for the outer edge of capability. If you were paying Fable prices mainly for reliable day-to-day agentic and coding work rather than for genuinely frontier-only tasks, Opus 5.5 is built to make that spend hard to justify.
What's next: Sonnet 5.5 and Haiku 5.5
Anthropic said Sonnet 5.5 and Haiku 5.5 will follow within weeks, extending the same efficiency push down the stack. Going by the established Claude 5-generation pricing (Sonnet 5 at $2/$10 and Haiku 4.5 as the current budget option), expect Sonnet 5.5 to stay the default 'daily driver' tier and Haiku 5.5 to target the highest-volume, lowest-latency use cases — though Anthropic hasn't published pricing or benchmarks for either yet, so treat any specific numbers you see elsewhere as unconfirmed until Anthropic posts them.
- Opus 5.5: $4/$20 per million tokens, live now, positioned as a cheaper near-Fable option for agentic and coding work.
- Sonnet 5.5: expected within weeks; likely to stay the mid-tier default for general coding, writing, and analysis.
- Haiku 5.5: expected within weeks; likely to target speed and cost over raw capability.
- All three would complete the 5.5 refresh across Anthropic's full public lineup, mirroring how the 5.1 refresh touched Fable and Mythos in June 2026.
If you're currently on Opus 5 for high-volume agentic workflows, the cache-read cut (60%, down to $0.20/M) is likely to matter more for your bill than the headline input/output price drop — benchmark your own long-context, tool-heavy tasks before assuming the published scores transfer directly.
01Is Claude Opus 5.5 better than Fable 5.1?
It depends on the task. Anthropic's own numbers show Opus 5.5 ahead on FrontierCode v1.1 Main and CursorBench 4.0, and claims a lead on Terminal-Bench 4.0, computer use, and chart recognition. Independent testing reportedly puts the Terminal-Bench gap much closer, and on Artificial Analysis's own Intelligence Index Fable 5.1 still trails Opus 5.5 only modestly (53 vs. 58), not decisively. On most everyday work, Anthropic frames the two as roughly comparable.
02How much cheaper is Opus 5.5 than Opus 5?
20% lower list pricing on input/output tokens ($4/$20 vs. $5/$25), a 60% cut to cached-token reads ($0.20/M), and an Anthropic-estimated 40% lower cost overall for typical workloads, plus 30%+ faster output.
03Is Opus 5.5 Anthropic's new flagship model?
No. Fable 5.1 (and the gated Mythos 5.1) remain Anthropic's top public tier. Opus 5.5 is a cheaper, faster model in the Opus line that closes much of the performance gap to Fable 5.1 on common tasks without replacing it at the top.
04When are Sonnet 5.5 and Haiku 5.5 launching?
Anthropic said on September 22, 2026 that both will arrive 'within weeks' of Opus 5.5, but hasn't given exact dates, pricing, or benchmarks yet.
05Where can I use Claude Opus 5.5?
It's available through the Anthropic API and on AWS, Google Cloud, and Microsoft Azure.
If you're trying to decide whether Opus 5.5, Fable 5.1, or something from another lab is the right fit for your workload, that's exactly the kind of side-by-side LumiChats is built for — you can chat with and compare multiple leading models in one place instead of juggling separate subscriptions and tabs to figure out which one actually earns its price.
