Anthropic released Opus 5.5 on September 22, 2026, with output tokens priced at $20 per million, 20% less than Opus 5, and an overall 40% reduction in costs on typical workloads, according to the company’s announcement. The model outperforms Fable 5.1 on most coding and knowledge work benchmarks while generating output more than 30% faster than its predecessor.
What changed in Opus 5.5?
Opus 5.5 arrives two months after Opus 5 launched on July 24, 2026. Anthropic describes it as a “major step up,” with early testers reporting large performance gains on complex, multi-step work. One tester completed a 680,000-line code migration in less than a day, work Anthropic says would have taken an engineering team weeks. A separate tester audited and fixed a 200,000-line codebase in under three hours, a task that took Opus 5 over 20 hours using 2.5x as many tokens.
The model also changes how it communicates. According to Anthropic, Opus 5.5 puts the most important information up front and avoids jargon, a direct response to feedback about Opus 5. Early testers noted the writing is “clearer and easier to follow” and that it “writes the way I do.”
How much cheaper and faster is Opus 5.5?
The price reductions apply across all token types, per Anthropic’s published pricing for September 2026:
- Input tokens: $4 per million (20% less than Opus 5)
- Output tokens: $20 per million (20% less than Opus 5)
- Cache reads (tokens retrieved from stored context, which make up the majority of agentic and coding work costs): $0.20 per million, 60% less than Opus 5
A Fast mode for Opus 5.5, available in Claude Code and the Claude Platform, delivers up to 2.5x speed at $8 per million input tokens and $40 per million output tokens.
In Anthropic’s own internal test, Opus 5.5 and Fable 5.1 were both asked to translate HAProxy (widely used software that balances web traffic loads across servers) from C into Rust. Both rewrites passed nearly all of HAProxy’s own regression tests (tests that verify a rewrite doesn’t break existing functionality), but Opus 5.5 finished in 9.5 hours versus 12 hours for Fable 5.1, and cost 51% less.
How does Opus 5.5 compare to competitors?
According to Anthropic’s September 2026 release data, Opus 5.5 leads on agentic coding, computer use, and knowledge work benchmarks. On FrontierCode, it beats GPT-6 Astra at roughly 20% of the cost per task. On Terminal Bench 4.0, it matches Astra for about 40% of the cost. On CursorBench, it beats GPT-5.6 Sol by 11 points for about a third of the cost.
Anthropic cautions that at these capability levels, benchmark margins have become “a less reliable guide to real-world differences.” The company notes the gap between Opus 5.5 and Fable 5.1 is narrower in practice than benchmark scores suggest, where Opus 5.5’s clear advantage is efficiency rather than raw capability.
What are the safety and pacing considerations?
Opus 5.5 is the first Anthropic model release since CEO Dario Amodei publicly called for pacing the frontier. “I have become convinced that fully addressing the risks requires even more prudence,” Amodei wrote in a post earlier in September 2026, “not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”
On Anthropic’s automated behavioral audit, which tests Claude across thousands of simulated scenarios, Opus 5.5 is the strongest-performing model the company has tested to date. It is less likely to take hard-to-reverse actions or act outside set boundaries, and more resistant to prompt injection (attempts by malicious content in the model’s input to override its instructions) than Opus 5.
Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity capabilities, Anthropic is deploying it with the same safeguards applied to Fable 5.1. Vetted organizations can apply to Anthropic’s Life Sciences Verification Program to use Opus 5.5 for biology research. Expansion of the Cyber Verification Program is expected in the coming weeks.
Pre-release evaluation was conducted by external organizations including METR and Frontier Design, consistent with Anthropic’s prior model releases.
What this means for AI-search visibility
The communication change in Opus 5.5 may be the most underreported part of this release for content teams. A model trained by default to front-load key information and avoid jargon produces output that structurally aligns with how AI answer engines, such as ChatGPT, Perplexity, and Google AI Overviews, retrieve and surface content.
Here is the practical implication. When those systems answer a user’s question, they pull from sources that are easy to parse: a direct answer near the top, clear structure, no buried lead. Opus 5.5 now writes this way by default. A team using it to draft or refine content may find the output already closer to AEO (Answer Engine Optimization, meaning content structured to be cited by AI rather than just ranked by search) standards without extra prompting or editing passes. That is a workflow efficiency that does not show up in benchmark tables.
The cost story also matters for content operations specifically. The 60% reduction in cache read pricing changes the economics of agentic publishing workflows, those where an AI runs repeated tasks across a large content library. Refreshing hundreds of pages, auditing internal links, or normalizing structured data across a site become meaningfully cheaper at $0.20 per million cache reads versus the prior rate. Teams running these workflows at volume should recalculate their cost models.
One signal worth watching separately: Anthropic’s stated pacing decision introduces a period of relative capability stability. If the company holds to slower frontier releases, the retrieval architectures inside AI search tools are less likely to change rapidly in the near term. That makes the structural content work done now, building citable, well-organized pages, more durable than during periods of fast model turnover. Whether pacing holds in practice is an open question, but the intent is explicit.
Sonnet 5.5 and Haiku 5.5 are expected in the coming weeks, with “similar performance improvements” per Anthropic’s announcement. The cost reductions across the full model family will lower the entry point for AI-assisted content workflows across different budget tiers.
Before upgrading your AI content pipeline to Opus 5.5
- Audit your current token consumption split: input, output, and cache reads. Cache reads see the largest price drop (60%), so pipelines heavy on cached context benefit most.
- Check whether your existing prompts rely on Opus 5’s verbosity or structure. Opus 5.5’s front-loaded, jargon-light output may change the shape of responses you have built workflows around.
- If you run unattended agentic tasks (long overnight jobs), review Anthropic’s notes on hard-to-reverse actions. Opus 5.5 is less likely to take them, but task design still matters.
- Verify whether your use case touches biology or cybersecurity. Access for those categories requires verification through Anthropic’s Life Sciences or Cyber Verification Programs.
- Test Fast mode separately if speed is a priority. At $8/$40 per million input/output tokens, it costs more per token but may reduce wall-clock time on latency-sensitive tasks.
- Compare costs on a representative sample of real tasks, not just token counts. Anthropic notes Opus 5.5 also uses fewer tokens per task, which compounds the per-token savings.
Sources: TechCrunch, September 22, 2026; Anthropic product announcement, September 22, 2026.
Lam Nguyen · Hingewise
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.