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 3 · 8 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 and with looser safety restrictions. It surpasses Fable 5 on several benchmarks while reducing the expected trigger frequency of safety classifiers by 85%, attempting to strike a new balance between capability, cost, and control.
Opus 5's API pricing is exactly the same as Opus 4.8: $5 per million input tokens and $25 per million output tokens. This pricing is half that of Fable 5 ($10 per million input tokens, $50 per million output tokens). Anthropic explicitly positions Opus 5 as a model that offers "near Fable 5 frontier intelligence at half the price."
Notably, Anthropic introduces a Fast mode that boosts speed by about 2.5x but doubles the price. Additionally, the model offers five "effort" settings (low, medium, high, xhigh, max), allowing users to trade off between performance and token consumption. Anthropic recommends using "low" or "medium" for most scenarios, but still suggests starting with "xhigh" for coding and agent tasks.
According to Anthropic's internal benchmarks, Opus 5 sets new highs on several evaluations. On Frontier-Bench v0.1, Opus 5's end-to-end coding score (43.3%) is more than double Opus 4.8 (21.1%) and surpasses Fable 5 (33.7%) and GPT-5.6 Sol (34.4%). On the knowledge work benchmark GDPval-AA v2, Opus 5's Elo rating (1,861) leads Fable 5 (1,747) and GPT-5.6 Sol (1,736).
The most striking result is on ARC-AGI-3, which tests a model's ability to solve novel problems. Opus 5 scores 30.2%, nearly four times that of second-place GPT-5.6 Sol (7.8%). However, this test lacks a comparison with Fable 5, and some observers believe this huge lead may not be reproducible in real-world use.
However, Opus 5 does not lead on all tests. On the agentic coding benchmark DeepSWE v1.1, GPT-5.6 Sol (72.7%) leads Fable 5 (69.7%) and Opus 5 (68.8%). On health and legal tasks, Fable 5 and Mythos 5 still perform better. Additionally, on cybersecurity tasks, Opus 5 approaches Mythos 5 in vulnerability discovery but lags significantly in exploit development.
Anthropic claims Opus 5 is its "most aligned model," with a composite misalignment score of 2.3, exhibiting the lowest rates of reckless or deceptive behavior and the strongest adherence to Claude's Constitution. The safety classifier trigger frequency is expected to be 85% lower than Fable 5, a figure based on Anthropic's projections and not yet independently verified.
In terms of cybersecurity guardrails, Opus 5 allows source code vulnerability scanning but prohibits binary vulnerability scanning, penetration testing, and exploit generation. Anthropic states that Opus 5 is stronger than Opus 4.8 on cybersecurity tasks but far weaker than Mythos 5. When requests are blocked on Claude.ai, Claude Code, and Claude Cowork, the default fallback is Opus 4.8, the same strategy as Fable 5.
Anthropic emphasizes that Opus 5 shows significant improvement in checking its own work and iterating. In one Frontier-Bench task, Opus 5 received a drawing of a machine part but could not directly view the drawing; it wrote its own computer vision pipeline to extract geometric shapes from raw pixels and reconstructed the complete machine part. Anthropic says other models failed to solve the task after five attempts.
In another case, Opus 5 fixed a real bug in a popular open-source package manager, identifying the root cause and fixing an edge case that the community patch had missed. Anthropic also mentions that an engineer at a trading firm used Opus 5 to build a market data feed for a new exchange in a single session, a task that previous models could not complete even with detailed plans.
This article is primarily based on Anthropic's official release and media reports. Benchmark scores, the 85% reduction in safety classifier trigger frequency, and other data are Anthropic's internal data or projections, lacking independent third-party verification. The significant lead on ARC-AGI-3 lacks a Fable 5 comparison, and its practical significance is questionable. Negative user feedback (e.g., accidental deletion of code repositories) has not been officially confirmed.
Opus 5 represents Anthropic's careful balancing act between performance, cost, and control: it delivers near-flagship performance at half the price of Fable 5 while significantly reducing safety intervention frequency. For most developers, Opus 5 may be a more practical choice than Fable 5, but it still lags behind competitors on some tasks, and its impressive ARC-AGI-3 score requires further validation.
Primary report
Primary source
Also covered by 5 sources