Anthropic has launched Claude Opus 5, a thoughtful and proactive AI model that approaches the frontier intelligence of Fable 5 at half the price. It achieves state-of-the-art results on coding and knowledge work evaluations, and scores three times higher than the next best model on ARC-AGI-3. Opus 5 is available today on all paid plans and the Claude API.
SynthePulse Insight · AI deep reading
Version 5 · 13 sources
On July 24, 2026, Anthropic released Claude Opus 5, positioning it as a model with performance close to the flagship Fable 5 but at half the price, while attracting a broader user base with looser safety classifiers and lower trigger frequency.
Opus 5 demonstrates strong performance across multiple benchmarks. On Frontier-Bench v0.1, Opus 5 scores 43.3%, more than double Opus 4.8 (21.1%), and surpasses Fable 5 (33.7%) and GPT-5.6 Sol (34.4%). On ARC-AGI-3, Opus 5 scores 30.2%, nearly four times that of second-place GPT-5.6 Sol (7.8%). However, these results are primarily from Anthropic internal data with limited independent verification, and ARC-AGI-3 lacks comparative data for Fable 5.
API pricing for Opus 5 matches Opus 4.8 ($5/M input tokens, $25/M output tokens), half that of Fable 5. Anthropic claims Opus 5 delivers better performance at the same or lower cost; for example, on CursorBench 3.2 at the highest effort setting, it trails Fable 5's peak score by only 0.5% at half the cost.
Opus 5 is described by Anthropic as its most aligned model to date, with a composite misalignment score of 2.3 and the lowest rates of reckless or deceptive behavior in automated behavior audits. Safety classifier trigger frequency is expected to be 85% lower than Fable 5, allowing vulnerability scanning in source code but prohibiting binary vulnerability scanning, penetration testing, and exploit generation. This design aims to balance safety and usability, making Opus 5 less restrictive than Fable 5.
In cybersecurity capabilities, Opus 5 is stronger than Opus 4.8 but lags behind Mythos 5. On OSS-Fuzz evaluations, Opus 5's success rate in identifying vulnerabilities is close to Mythos 5, but it is far behind in developing exploit programs. Biosecurity protections continue the previous generation's design, and Opus 5 still has clear limitations on long-horizon autonomous research tasks.
Despite strong performance in many areas, Opus 5 lags behind competitors on certain tasks. On DeepSWE v1.1, GPT-5.6 Sol leads Opus 5. On health and legal tasks, Fable 5 and Mythos 5 remain ahead. Additionally, Fast mode, while offering ~2.5x speed, doubles the price, which may limit its practical application.
User feedback includes negative experiences (e.g., accidentally deleting code repositories), but these have not been officially confirmed. Anthropic emphasizes that Opus 5's general access has no data retention requirements, but requests blocked by the safety classifier default to being handled by Opus 4.8.
This article is primarily based on official Anthropic announcements and media reports. Some benchmark results are Anthropic internal data with limited independent verification. The 85% reduction in safety classifier trigger frequency is an Anthropic expectation, with actual effectiveness pending verification. The significant lead on ARC-AGI-3 may not be replicable in real-world use.
Opus 5 offers near-flagship performance at half the price with looser safety controls, making it Anthropic's most cost-effective model currently. However, it does not lead in all tasks, and some advantages rely on internal data requiring independent verification.
Primary report
Primary source
Also covered by 14 sources