AI News Feed
Market watch
Mistral Denies New Security Breach as Code for Sale Echoes May LeakLCD Panel Price Increases Threaten Black Friday TV Deals as Samsung Trims QLED OutputMicrosoft Disrupts EvilTokens, an AI-Assisted Scam Platform Tied to 12,000 Compromised AccountsBernstein Stays Bullish on Nvidia and Broadcom, Sees 58% and 75% UpsideRabbit rolls out OS3 AI agent that doesn't need R1 hardwareAnthropic launches Claude Opus 5.5 with faster output and about 40% lower running costs
Mistral Denies New Security Breach as Code for Sale Echoes May LeakLCD Panel Price Increases Threaten Black Friday TV Deals as Samsung Trims QLED OutputMicrosoft Disrupts EvilTokens, an AI-Assisted Scam Platform Tied to 12,000 Compromised AccountsBernstein Stays Bullish on Nvidia and Broadcom, Sees 58% and 75% UpsideRabbit rolls out OS3 AI agent that doesn't need R1 hardwareAnthropic launches Claude Opus 5.5 with faster output and about 40% lower running costs
Large Language Models

Anthropic launches Claude Opus 5.5 with faster output and about 40% lower running costs

Anthropic released Claude Opus 5.5 on Sept. 22, promising faster, cheaper and more accurate output than Opus 5, along with stronger safety safeguards.

AI News Feed · 2026-09-23 06:38

The price change applies to API usage. Input tokens fell to $4 per million from $5, and output tokens to $20 per million from $25. Cache reads cost $0.20 per million tokens, down from $0.50, while cache writes cost $5 per million tokens, down from $6.25. Anthropic said that when combined with better token efficiency, typical workloads should run about 40 percent cheaper than on Opus 5.

In benchmarks covering agentic coding, knowledge work, multidisciplinary reasoning, computer use and visual chart recognition, Anthropic said Opus 5.5 leads Fable 5.1, Opus 5, GPT-6 Astra and GPT-5.6 Sol. It placed second, behind GPT-6 Astra, in business workflows and agentic scientific research.

Anthropic also cited early testers to back the benchmark figures. One tester audited and fixed a 200,000-line codebase in under three hours, a task Anthropic said took Opus 5 more than 20 hours and 2.5 times as many tokens. In a test that translated software from C into Rust, Opus 5.5 finished in 9.5 hours, compared with 12 hours for Fable 5.1. A tester at GitHub found the model solved more terminal tasks than Opus 5 in less than half the steps, while a tester at Spotify said it completed tasks faster and more cheaply than earlier models. Anthropic also reported results from early testers at Clio, Lovable and Quantum.

Anthropic said the new model also addresses complaints about how Opus 5 communicated. A Ramp software engineer wrote that verbose, hard-to-follow output had been the biggest frustration with frontier models and that Opus 5.5 fixes it, adding that it writes like a good colleague and follows the company's writing rules. A principal engineer at the Chicago Trading Company said the writing is easy to follow and more coherent than Opus 5's.

On safety, Anthropic said Opus 5.5 includes a native classifier that screens every action before it runs. The company said the model matches or beats Opus 5 at mitigating prompt injection attacks in coding, tool use, computer use and web browsing, and that on a benchmark run by the AI security firm Gray Swan it tied Fable 5.1 for the lowest prompt injection success rate of any model tested.

Opus 5.5 is not limited to developers using the API. It is available to regular users, with five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans. Anthropic said usage limits for those subscribers are being increased.

Editor's Summary

Anthropic released Claude Opus 5.5 on Sept. 22, cutting API prices and claiming more than 30 percent faster output and roughly 40 percent lower running costs than Opus 5. The company says the model leads rivals including Fable 5.1, GPT-6 Astra and GPT-5.6 Sol in several benchmarks and adds stronger defenses against prompt injection. Access was extended to regular subscribers on Pro, Max, Team and Enterprise plans.