Anthropic Releases Claude Opus 5 with Frontier Benchmark Leadership and Alignment Claims

Anthropic released Claude Opus 5 on July 24, 2026, claiming it leads all competitors on Frontier-Bench v0.1 and scores three times higher than the next-best model on ARC-AGI 3, while more than doubling Opus 4.8's per-task performance at equal or lower cost [1]. Anthropic also designates Opus 5 its most aligned model to date, citing reduced deceptive behavior and stronger resistance to manipulation [1]. This follows the June 30, 2026 release of Claude Sonnet 5, positioned as an agentic model approaching Opus 4.8 performance at a lower price tier, with introductory pricing through August 31, 2026 [2]. Both models are available through the Claude API and in Claude Code; Sonnet 5 is now the default for Free and Pro consumer plans [2].

If the benchmark figures hold under independent scrutiny, Opus 5 represents a real cost-efficiency advance at the frontier — significantly better performance at unchanged Opus 4.8 prices. Anthropic's claim that safety and capability improved together matters because that relationship is empirically contested across the field, and Anthropic is staking its brand positioning on it.

Anthropic released Claude Opus 5 on July 24, 2026, with claims of broad benchmark leadership. The company reports Opus 5 tops Frontier-Bench v0.1 across all tested models, scores three times higher than the next-best model on ARC-AGI 3, and achieves roughly 1.5 times the pass rate of the next-best model on Zapier AutomationBench at equivalent cost [1]. Underlying these benchmark figures is a cost-efficiency claim: Opus 5 reportedly more than doubles Opus 4.8's performance per task while carrying identical list pricing — $5 per million input tokens and $25 per million output tokens — with a Fast mode running at approximately 2.5 times default speed at twice that base price [1].

Alongside the capability claims, Anthropic's Opus 5 announcement includes detailed safety disclosures. The company's behavioral audits rate Opus 5 as its most aligned model, with the lowest measured rates of deceptive behavior and better adherence to Claude's Constitution than Opus 4.8, Sonnet 5, or a third model named Fable 5 [1]. On dual-use risks, Anthropic states Opus 5 does not advance the frontier in dual-use biology or offensive cybersecurity — but it notes that a separate model named Mythos 5 remains ahead of Opus 5 on exploitation of discovered vulnerabilities, without explaining Mythos 5's deployment status or access controls [1]. This reference introduces two model names (Fable 5 and Mythos 5) that had not previously appeared in public Anthropic communications.

Opus 5's release came roughly three weeks after Claude Sonnet 5, which Anthropic released on June 30, 2026 [2]. Sonnet 5 was positioned as the most agentic Sonnet model to date, capable of autonomous multi-step tasks at a level that previously required larger models. Anthropic reports Sonnet 5 shows lower hallucination, sycophancy, and undesirable behavior than Sonnet 4.6, while pricing it at $2 per million input tokens with an introductory rate lasting through August 31, 2026 [2]. Sonnet 5 replaced prior defaults on Free and Pro consumer plans and is available in Claude Code. Together the two releases update Anthropic's full public model stack.

In the weeks before Opus 5's announcement, third-party blogs and YouTube channels published speculation about its release timing and expected capabilities, but those items contained no verifiable claims [3][4][5]. The official July 24 announcement displaced that speculation with a system card covering benchmarks, pricing structure, and safety audit methodology.