Anthropic slashes Claude Opus cost by 40% as speed jumps 30%

Anthropic has pulled back the curtain on its newest flagship model, and the headline number is hard to ignore: Claude Opus 5.5 runs at roughly 40% lower cost than its predecessor while outperforming it on coding and knowledge benchmarks. The Anthropic Claude Opus line has always positioned itself as the premium tier of the company’s three-model lineup, sitting above Sonnet and Haiku, and this release suggests the company is trying to make that premium tier less expensive to run without giving up the performance edge that justified the higher price in the first place.

Key takeaways

• Claude Opus 5.5 cuts operating costs by about 40% compared with Opus 5, according to Anthropic.
• Input token pricing dropped from $5 to $4 per million tokens, and output pricing fell from $25 to $20 per million tokens.
• Output generation is over 30% faster than Opus 5, with a high-speed mode running up to 2.5 times standard velocity.
• The model scored 66.4% on Terminal-Bench 4.0, ahead of GPT-6 Astra’s 57.9% and GPT-5.6 Sol’s 37.3%.
• Opus 5.5 underwent external safety evaluations from METR and Frontier Design and showed 85% fewer attempts to bypass containment controls.

Anthropic launches Claude Opus 5.5 with reduced operating costs

Anthropic released Opus 5.5 on Tuesday, positioning it as the newest entry in what the company calls the Claude 5.5 series. According to Anthropic, the model delivers performance that rivals its most premium offerings while trimming running costs by roughly 40% relative to Opus 5. That combination — better output at a lower price — is the core pitch behind this Anthropic Claude Opus update, and it’s the kind of trade-off that tends to matter most to businesses running high-volume workloads rather than casual users.

Cost reductions in token pricing and operational efficiency

The pricing changes are concrete. Input tokens for Opus 5.5 now cost $4 per million, down from $5 for Opus 5. Output tokens dropped from $25 to $20 per million. Cache-read costs fell even more sharply in percentage terms, from $0.50 to $0.20 per million tokens. Anthropic says the model also achieves improved token efficiency across a wide range of tasks, which compounds with the headline price cuts to produce that overall 40% reduction for typical operations. Notably, TechCrunch reported that the drop in compute required to serve the model reflects a genuine efficiency gain rather than just a pricing adjustment layered on top of the same underlying system.

Introduction of high-speed mode with premium pricing

Anthropic is also rolling out a high-speed mode through Claude Code and the Claude Platform, letting users pay a premium for processing speeds up to 2.5 times the standard rate. That mode costs more — $8 per million input tokens and $40 per million output tokens — but it’s aimed at workflows where turnaround time outweighs cost sensitivity. Alongside this, subscription tiers including Pro, Max, Team, and seat-based Enterprise plans are getting expanded five-hour usage caps, plus a rate-limit reset feature members can save for later use.

Performance enhancements in coding and knowledge tasks

Beyond price, Anthropic is framing Opus 5.5 as a genuine capability jump, particularly for software engineering and professional knowledge work. The company reports output generation speeds more than 30% faster than Opus 5, a difference that becomes especially visible on long-running coding tasks.

Faster output generation and benchmark results

Anthropic says the model handles extended software engineering workflows — migrations, code reviews, debugging — with notably less friction. In one internal test, Opus 5.5 reportedly completed a 200,000-line codebase evaluation in under three hours, compared with more than 20 hours for Opus 5. On Anthropic’s Terminal-Bench 4.0 benchmark, Opus 5.5 scored 66.4%, ahead of GPT-6 Astra’s 57.9% and GPT-5.6 Sol’s 37.3%. Anthropic itself has acknowledged that benchmark outcomes can shift depending on model configuration and testing method, and that real-world performance gaps may be narrower than the raw scores suggest. TechCrunch also noted that Anthropic claims Opus 5.5 outpaces its own larger Fable model on several benchmarks and completed informal tasks that Fable could not finish — a notable claim given Fable’s position as a more resource-intensive system.

On the GDPval-AA v2.1 benchmark, which measures professional competency across 44 occupations, Opus 5.5 registered 1846 Elo, surpassing Anthropic’s earlier releases. This matters because it signals the company is chasing more than raw coding speed — it’s trying to prove the model holds up in white-collar, document-heavy work too.

Enterprise user feedback and professional task improvements

Anthropic says early enterprise adopters — including GitHub, Deloitte, and Stripe — have reported improvements in software creation, analytical tasks, and written communication. Those are company-supplied assessments rather than independently verified results across every deployment scenario, but they still point to where Anthropic expects the biggest commercial payoff: teams that lean on AI for both code and business documentation. TechCrunch also flagged a subtler change in Opus 5.5’s behavior — the model reportedly uses less jargon and is more likely to put the most important information at the start of its responses, a shift aimed at making outputs easier to scan quickly.

Enhanced safety and security measures in Claude Opus 5.5

Safety testing remains a central part of how Anthropic frames this release, and that’s not incidental. The launch arrives shortly after CEO Dario Amodei publicly argued for slowing the pace of frontier AI capability gains so that safety infrastructure has time to catch up — a position he laid out in a recent post, writing that “fully addressing the risks requires even more prudence… not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”

External safety evaluations and internal assessments

Opus 5.5 went through pre-launch evaluation from outside organizations, including METR and Frontier Design, before deployment. Anthropic reports the model posted its best scores yet on the company’s own automated behavioral assessments. TechCrunch reported that this safety training was broadly similar in structure to previous releases, though Anthropic said more advanced training and evaluation systems — including improved security and monitoring tools — are already being prepared for future models.

Improved resistance to containment circumvention and prompt injection

Anthropic states that Opus 5.5 showed approximately 85% fewer attempts to circumvent containment parameters in specialized testing, compared with both Opus 5 and Claude Mythos 5.1. The company also reports improved resistance to prompt-injection attempts. Because the model shows strong proficiency in cybersecurity and biological domains, Anthropic has built in additional protective mechanisms for those areas — comparable to the safeguards applied to its Fable model, which restrict how the system can be used to find exploits in compiled programs or assist with biological weapons development. Anthropic is also expanding verification programs that give vetted researchers and cybersecurity professionals broader access for approved projects.

Availability and upcoming releases in the Claude 5.5 series

Opus 5.5 is available now directly through Anthropic, as well as via Amazon Web Services, Google Cloud, and Microsoft Azure — giving enterprise customers multiple paths to deploy it inside existing cloud infrastructure. That multi-cloud availability matters for businesses that have already built workflows around one of those providers, since it removes a potential barrier to adoption.

Anticipated launches of Claude Sonnet 5.5 and Claude Haiku 5.5

Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 — the middle and entry tiers of the lineup — are expected to arrive within the next several weeks, with similar performance gains. The Opus 5.5 launch comes roughly two months after Opus 5 debuted in July, a turnaround that shows how quickly Anthropic is iterating even while publicly emphasizing a more cautious approach to raw capability gains. Once the Sonnet and Haiku updates land, the full Anthropic Claude Opus Sonnet Haiku lineup will reflect the same cost and speed improvements introduced with this release, giving developers a consistent set of trade-offs to choose from across price points.

FAQ

What are the cost improvements of Claude Opus 5.5 compared to its predecessor?

Claude Opus 5.5 reduces operating costs by roughly 40%, with input token pricing dropping from $5 to $4 per million tokens and output token pricing from $25 to $20 per million.

How does Claude Opus 5.5 perform in coding and professional tasks?

It offers improved coding and knowledge processing with output generation speeds over 30% faster than Opus 5 and achieves higher benchmark scores such as 66.4% on Terminal-Bench 4.0.

What safety enhancements does Claude Opus 5.5 feature?

Opus 5.5 completed external safety evaluations, scored highest on internal automated behavioral assessments, and shows approximately 85% fewer containment circumvention attempts along with improved resistance to prompt injection.

Which enterprises are using Claude Opus 5.5?

Enterprise users include GitHub, Deloitte, and Stripe, who reported enhancements in software development, analytical tasks, and written communication.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.