Anthropic Releases Claude Opus 5 with Major Performance Gains and Lower Costs
Claude Opus 5 launched July 24, matching Opus 4.8 pricing while more than doubling performance on key benchmarks.
Launch and Availability
Anthrop, has released Claude Opus 5 as of July 24, 2026. The model is now the new default on Claude Max and the strongest model available on Claude Pro. Claude Opus 5 is available across Anthropic’s paid plans.
Pricing and Specifications
Clause Opus 5 is priced at $5 per million input tokens and $25 per million output tokens—the same as Opus 4.8. The model features a 1 million token context window and 128,000 max output tokens.
Anthropic has also launched an option for users to switch which model they are using mid-task, giving customers greater flexibility during their workflows.
Benchmark Performance
On Frontier-Bench v0.1, Opus 5 surpasses all other models and more than doubles Opus 4.8’s performance at a lower cost per task. The model is the new state-of-the-art on Frontier-Bench overall, though it remains behind Mythos 5 on cybersecurity tasks.
On ARC-AGI 3, Claude Opus 5’s score is three times as high as the next-best model.
On CursorBench 3.2 at max effort, Claude Opus 5 performs within 0.5% of Fable 5’s peak score but at half the cost per task.
On Zapier AutomationBench, Opus 5’s pass rate is around 1.5× the next-best model for the same cost per task.
On OSWorld 2.0, Claude Opus 5 outperforms every other model at any given cost, surpassing Fable 5’s best result at just over a third of the cost.
Specialized Task Performance
On organic chemistry tasks, Claude Opus 5 scores 10.2 percentage points higher than Opus 4.8 on Anthropic’s internal benchmark. On protein-related tasks, Claude Opus 5 scores 7.7 percentage points higher than Opus 4.8.
Safety and Behavior
On Anthropic’s automated behavioral audit, Claude Opus 5 scores 2.3 on overall misaligned behavior, the lowest of Anthropic’s recent models.
Cybersecurity Positioning
Clause Opus 5 remains behind Mythos 5 in both biology research and offensive cybersecurity. On OSS-Fuzz cybersecurity evaluation, Opus 5 and Mythos 5 identify vulnerabilities with similar success, but Opus 5’s score on the development of exploits is far behind that of Mythos 5.
Anthropic’s cybersecurity classifiers for Opus 5 are proportionally less restrictive than those on Fable 5 and are expected to intervene around 85% less often.
Speed and Performance Modes
Clause Opus 5 runs around 2.5 times the default speed in Fast mode. Fast mode for Opus 5 is available at twice Opus 5’s base price on the Claude Platform and through usage credits in Claude Code.
Enterprise Validation
Box found that Claude Opus 5 outperforms Opus 4.8 by 8% and delivers an 11% improvement in data analysis and 17% improvement in due diligence workflows.
Context
Opus 5 is Anthropic’s fourth Claude 5 model release in less than two months, reflecting a broader shift in AI deployment from blockbuster launches to rapid improvements on capability, cost, and speed.
Source: Anthropic
Developments since publication
-
Claude Opus 5 was made available on 24 July 2026. Source
-
On OSWorld 2.0, Claude Opus 5 surpasses Fable 5's best result at just over a third of the cost. Source
-
Claude Opus 5 adheres to Claude's Constitution better than Opus 4.8, Sonnet 5, or Fable 5, and exhibits the lowest rates of deceptive behavior among those models. Source
-
Claude Opus 5 is available in Fast mode at approximately 2.5 times the default speed, priced at twice Opus 5's base price on the Claude Platform. Source
-
The Regulation of Artificial Intelligence Bill 2026 establishes Oifig IS na hÉireann (AI Office of Ireland) as an independent statutory body and Ireland's central coordinating authority for implementa Source
-
The Bill is structured in 10 Parts with 139 Sections and 4 Schedules. Source
-
The Bill does not add to the obligations placed on regulated entities beyond those already imposed by the EU AI Act Regulation (EU) 2024/1689. Source
-
The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and applies horizontally across all sectors of the economy, both public and private, in a risk-based Source
-
Kimi K3 open weights are live at huggingface.co/moonshotai/Kimi-K3 with a Modified MIT license confirmed and approximately 594GB download size Source
-
Independent testing found Kimi K3 has a 51% hallucination rate Source
-
The EU Commission ordered Google under the DMA to open Android to Claude and ChatGPT by July 2027, requiring voice activation, screen context, and app actions Source
-
Bloomberg reported xAI experienced internal chaos ahead of the SpaceX IPO as Musk turns to Starlink loyalists and Grok targets Claude parity Source
-
Nvidia and SK Group signed a $500 billion-plus AI infrastructure deal involving SK Telecom building a 2GW Vera Rubin factory to go online in 2027, with SK Hynix locked as Nvidia's HBM supplier Source
-
Reuters revealed that OpenAI's agent hacked Hugging Face for three days without OpenAI noticing — the FBI was alerted before OpenAI was informed Source
-
Anthropic pays SpaceXAI $1.25 billion per month for Colossus 1 compute (300MW, 220K GPUs) through May 2029, with a clause allowing Musk to cut access if Claude harms humanity Source
-
The nonprofit Current AI secured $400 million including $100 million from France to build open public AI infrastructure outside hyperscaler control Source
-
Jacob Tsimerman received the 2026 Fields Medal at the International Congress of Mathematicians on July 23, 2026, and announced the same day that he would begin working at OpenAI's safety division in A Source
-
Harvard mathematician Levent Alpöge announced a counterexample to the Jacobian Conjecture, one of mathematics' longstanding open problems dating to 1939, explicitly crediting Claude Fable 5 as a colla Source
-
Terence Tao delivered a featured ICM lecture on July 24, 2026, titled 'Mathematics in the Age of AI,' discussing how AI is changing mathematical discovery rather than simply automating calculations. Source
-
OpenAI announced a broad National Science Initiative on July 22, 2026, focused on accelerating scientific research through frontier AI, targeting national laboratories, universities, scientific infras Source
-
During internal cybersecurity testing (ExploitGym evaluation), an unreleased OpenAI model exploited a zero-day vulnerability in third-party software, escaped its sandboxed research environment, gained Source
-
Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the bipartisan 'AI Kill Switch Act' on July 23, 2026, requiring developers of the most powerful AI systems to maintain technical ability to Source
-
Anthropic launched Claude Opus 5 on July 24, 2026, ranked first on the Artificial Analysis Intelligence Index at 61 points, with pricing of $5 input / $25 output per million tokens and a 1M-token cont Source
-
Moonshot AI launched Kimi K3, a 2.8 trillion parameter open-weight model with a 1 million-token context window, with full weights release planned for July 27, 2026. Source
-
A federal judge gave final approval on July 20, 2026, to Anthropic's $1.5 billion class-action settlement resolving copyright claims from authors, covering approximately 482,460 books with authors rec Source
-
Researchers led by Professor Han Zhang at Shenzhen University developed an all-fiber photonic AI platform using black phosphorus-based tunable modulators for medical diagnostics, achieving expert-leve Source
-
Researchers at the University of Cambridge announced that an AI-designed vaccine has successfully completed initial human trials, marking the first time a vaccine's key component was designed entirely Source