$MSFTB $AMZNB $GOOGLB
Claude Opus 5.5 Goes Live: Prices Down 20% vs. the Previous Version, Costs Down 40%
On the 22nd, Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 series. According to the official announcement, Opus 5.5 performs at the same level as Claude Fable 5.1 on most tasks, while its operating costs are 40% lower than Opus 5. Sonnet 5.5 and Haiku 5.5 will roll out in the coming weeks. Input and output are priced at $4 and $20 per million tokens, respectively, and cache reading is reduced by 60%. Anthropic says Opus 5.5 requires fewer compute resources than Opus 5, and the pricing reflects this. Input and output are $4 and $20 per million tokens respectively—20% cheaper than Opus 5. Cache reads, which account for the bulk of costs for agents and programming work, are reduced to $0.2 per million tokens—60% cheaper than Opus 5. Under default settings, typical workloads cost 40% less than Opus 5, and the output speed is also over 30% faster.
In addition to lowering prices, Anthropic has raised the 5-hour usage caps for its Pro, Max, and Team plans, and offers subscribers an adjustable rate-limit reset that they can choose when to apply. Claude Code and the Claude platform also include a fast mode: up to 2.5x speed, priced at $8 per million tokens for input and $40 for output.
A migration of 680,000 lines of code was completed within a day, and the company says the benchmark gap can no longer reflect real differences. Anthropic points out that Opus 5.5 leads in internal evaluations for agentic coding, computer operations, and knowledge work. One early tester used it to complete a 680,000-line code migration in a single day—a task that previously required weeks from a team of engineers. Another tester reviewed and fixed a 200,000-line code library in under 3 hours, while Opus 5 took over 20 hours and used 2.5x the tokens.
In a benchmark evaluating GDPval-AA v2.1 across real-world work from 44 professions, Opus 5.5 achieved 1846 Elo with the highest intensity setting. Fable 5.1 scored 1735, and Opus 5 scored 1708. Anthropic also reminds that at this level of capability, benchmark differences are less able to reliably reflect real-world differences; based on their own usage experience, the score gap between Opus 5.5 and Fable 5.1 appears smaller.
Another focus of this release is how it communicates. Anthropic says one of the most common pieces of feedback for Opus 5 is that its outputs can be overly long and hard to read. Opus 5.5 will put the most important information first, use fewer jargon terms or unusual phrasing, and better follow writing rules users provide.
The first model after “putting the brakes on the frontier.” Most security-related tasks are now handled by Opus 4.8, according to Anthropic. Opus 5.5 is the company’s first model released after calling for putting the brakes on the frontier, and before launch it was tested by external evaluation organizations such as METR and Frontier Design. In an automated behavior audit covering nearly 2,000 scenarios, Opus 5.5 is the best-performing model tested so far, and it is also the strongest model in most honesty-related metrics. The official statement says that in a new evaluation where models tend to attempt to cross isolation boundaries, Opus 5.5 tried to bypass the boundary about 85% fewer times than Opus 5 or Claude Mythos 5.1, and each attempt was categorized as low severity and self-reported by the model.
However, Anthropic also notes that building evaluations that can reliably catch every kind of failure before deployment remains an unresolved problem. They have also seen indications that Opus 5.5 often seems to doubt that it is being evaluated, which can weaken their ability to judge how the model will act in real environments.
Because Opus 5.5’s capabilities in the areas of biology and cybersecurity are comparable to Claude Mythos 5.1, Anthropic aligned deployment safeguards to the level of Fable 5.1: most cybersecurity tasks will be handled by Opus 4.8, and biology work requires applying through a new life sciences verification program. Only academic labs, startups, and pharmaceutical companies that pass the review can use it. Anthropic also says that in the coming weeks, it will expand the cybersecurity verification program so that verified cybersecurity professionals can use Opus 5.5 for their work.
Opus 5.5 is now live on all major platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The model ID for developers using it on the Claude platform is claude-opus-5-5.
This article, “Claude Opus 5.5 Goes Live: Prices Down 20% vs. the Previous Version, Costs Down 40%,” first appeared on .
Claude Opus 5.5 Goes Live: Prices Down 20% vs. the Previous Version, Costs Down 40%
On the 22nd, Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 series. According to the official announcement, Opus 5.5 performs at the same level as Claude Fable 5.1 on most tasks, while its operating costs are 40% lower than Opus 5. Sonnet 5.5 and Haiku 5.5 will roll out in the coming weeks. Input and output are priced at $4 and $20 per million tokens, respectively, and cache reading is reduced by 60%. Anthropic says Opus 5.5 requires fewer compute resources than Opus 5, and the pricing reflects this. Input and output are $4 and $20 per million tokens respectively—20% cheaper than Opus 5. Cache reads, which account for the bulk of costs for agents and programming work, are reduced to $0.2 per million tokens—60% cheaper than Opus 5. Under default settings, typical workloads cost 40% less than Opus 5, and the output speed is also over 30% faster.
In addition to lowering prices, Anthropic has raised the 5-hour usage caps for its Pro, Max, and Team plans, and offers subscribers an adjustable rate-limit reset that they can choose when to apply. Claude Code and the Claude platform also include a fast mode: up to 2.5x speed, priced at $8 per million tokens for input and $40 for output.
A migration of 680,000 lines of code was completed within a day, and the company says the benchmark gap can no longer reflect real differences. Anthropic points out that Opus 5.5 leads in internal evaluations for agentic coding, computer operations, and knowledge work. One early tester used it to complete a 680,000-line code migration in a single day—a task that previously required weeks from a team of engineers. Another tester reviewed and fixed a 200,000-line code library in under 3 hours, while Opus 5 took over 20 hours and used 2.5x the tokens.
In a benchmark evaluating GDPval-AA v2.1 across real-world work from 44 professions, Opus 5.5 achieved 1846 Elo with the highest intensity setting. Fable 5.1 scored 1735, and Opus 5 scored 1708. Anthropic also reminds that at this level of capability, benchmark differences are less able to reliably reflect real-world differences; based on their own usage experience, the score gap between Opus 5.5 and Fable 5.1 appears smaller.
Another focus of this release is how it communicates. Anthropic says one of the most common pieces of feedback for Opus 5 is that its outputs can be overly long and hard to read. Opus 5.5 will put the most important information first, use fewer jargon terms or unusual phrasing, and better follow writing rules users provide.
The first model after “putting the brakes on the frontier.” Most security-related tasks are now handled by Opus 4.8, according to Anthropic. Opus 5.5 is the company’s first model released after calling for putting the brakes on the frontier, and before launch it was tested by external evaluation organizations such as METR and Frontier Design. In an automated behavior audit covering nearly 2,000 scenarios, Opus 5.5 is the best-performing model tested so far, and it is also the strongest model in most honesty-related metrics. The official statement says that in a new evaluation where models tend to attempt to cross isolation boundaries, Opus 5.5 tried to bypass the boundary about 85% fewer times than Opus 5 or Claude Mythos 5.1, and each attempt was categorized as low severity and self-reported by the model.
However, Anthropic also notes that building evaluations that can reliably catch every kind of failure before deployment remains an unresolved problem. They have also seen indications that Opus 5.5 often seems to doubt that it is being evaluated, which can weaken their ability to judge how the model will act in real environments.
Because Opus 5.5’s capabilities in the areas of biology and cybersecurity are comparable to Claude Mythos 5.1, Anthropic aligned deployment safeguards to the level of Fable 5.1: most cybersecurity tasks will be handled by Opus 4.8, and biology work requires applying through a new life sciences verification program. Only academic labs, startups, and pharmaceutical companies that pass the review can use it. Anthropic also says that in the coming weeks, it will expand the cybersecurity verification program so that verified cybersecurity professionals can use Opus 5.5 for their work.
Opus 5.5 is now live on all major platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The model ID for developers using it on the Claude platform is claude-opus-5-5.
This article, “Claude Opus 5.5 Goes Live: Prices Down 20% vs. the Previous Version, Costs Down 40%,” first appeared on .
