Anthropic has released Claude Opus 5, a new AI model designed for everyday business needs that delivers near-frontier intelligence at half the cost of its flagship Claude Fable 5. Priced identically to Claude Opus 4.8 at $5 per million input tokens, the model introduces an adjustable effort setting to optimize token efficiency. Early enterprise users report significant performance gains and token reductions. While Claude Opus 5 excels in self-verification and coding benchmarks, Anthropic has intentionally restricted its offensive cyber capabilities, keeping it behind Mythos 5.
Claude Opus 5 launch
- ▪Claude Opus 5 becomes the new default model on Claude Max and the strongest model available on Claude Pro
- ▪Anthropic released Claude Opus 5 on July 24, 2026, as a new AI model geared toward everyday business needs
- ▪Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, which is unchanged from its predecessor, Claude Opus 4.8
Benchmark performance results
- ▪On the OSWorld 2.0 computer-use benchmark, Claude Opus 5 outperformed Claude Fable 5's best result at just over one-third of the cost
- ▪On the ARC-AGI 3 novel problem-solving evaluation, Claude Opus 5 scored three times as high as the next best model
- ▪On the Frontier-Bench v0.1 agentic terminal coding benchmark, Claude Opus 5 scored 43.3 percent, compared to 18.7 percent for Claude Opus 4.8 and 33.7 percent for Claude Fable 5
- ▪Claude Opus 5 scored 10.2 percentage points higher than Claude Opus 4.8 on Anthropic's internal organic chemistry benchmark
Token efficiency economics
- ▪Fundamental Research Lab reported that Claude Opus 5 averaged nine percentage points higher accuracy on financial-modeling tasks while using one-third fewer turns and 60 percent less time
- ▪Claude Opus 5 features an adjustable effort setting that allows users to toggle how much effort the model expends to balance cost and capability
- ▪Legal AI company Harvey reported that Claude Opus 5 achieved similar performance to Claude Opus 4.8's maximum-reasoning mode while generating 26 percent fewer tokens on average
Self-verifying agent capabilities
- ▪Stripe software engineer Cristian Rivera reported that Claude Opus 5 successfully managed development environments over a weekend, building its own monitor and driving each box autonomously
- ▪Claude Opus 5 is designed to verify its own work, recover from errors without intervention, and iterate until it succeeds
- ▪In a Frontier-Bench task where it could not view a drawing, Claude Opus 5 wrote its own computer vision pipeline to extract geometry from raw pixels and rebuild a 3D model
Safety architecture design
- ▪On the OSS-Fuzz evaluation, Claude Opus 5 identified software vulnerabilities at a 79.4 percent rate but succeeded in developing exploits in only 4 challenges
- ▪Anthropic intentionally avoided training Claude Opus 5 on cyber tasks, leaving it behind Mythos 5 in offensive cybersecurity and biology research
- ▪When a safety classifier declines a request in Claude.ai, Claude Code, or Claude Cowork, the system automatically falls back to Claude Opus 4.8 by default
- ▪Claude Opus 5 scored 2.3 on overall misaligned behavior in Anthropic's automated behavioral audit, making it the company's most aligned model to date
Anthropic valuation momentum
- ▪Anthropic's annualized revenue grew from about $1 billion at the end of 2024 to a projected $9 billion by the end of 2025
- ▪Anthropic has secured a reported $30 billion Azure compute deal alongside infrastructure partnerships with Google Cloud and Nvidia
- ▪Anthropic was valued at approximately $380 billion in a February 2026 funding round
Story comments
Loading comments…