Overview
Anthropicshipped Claude Opus 5 on July 24, 2026, positioning the model as a near-frontier alternative to its most capable commercially available system, Claude Fable 5, at half the application programming interface (API) price. <cite index="4-1,4-2">The release delivers performance close to Fable 5 at half the API cost</cite>, a pricing structure that Anthropic framed as making frontier-grade capabilities viable for everyday enterprise use without budget constraints.
Pricing and Availability
<cite index="3-1">Opus 5 is priced at $5 per million input tokens and $25 per million output tokens — the same as its predecessor Opus 4.8 and half the rate of Fable 5.</cite> <cite index="7-4">Fable 5, launched in June, is priced at $10 per million input tokens and $50 per million output tokens.</cite> <cite index="9-15">In roughly a year, Opus-tier pricing fell 3×, and Opus 5 more than doubles Opus 4.8's Frontier-Bench score on top of that.</cite>
<cite index="6-5">Developers can access it through the Claude API using the identifier `claude-opus-5`, while the model is now the default on Claude Max and the strongest model available through Claude Pro.</cite> An optional Fast mode runs the model at approximately 2.5 times standard speed but <cite index="9-12,9-13">costs double the base rate — $10 per million input tokens and $50 per million output tokens.</cite>
<cite index="22-7">Amid growing concerns from enterprise customers about expensive AI bills, Opus 5 ships with a feature enabling users to toggle how much effort — low, medium, or high — the model expends completing a task or answering a prompt.</cite>
Benchmark Performance
On the key agentic coding evaluation Frontier-Bench v0.1, <cite index="4-4">internal benchmarks show Opus 5 outperforms Fable 5, recording a 43.3% score versus Fable 5's 33.7%.</cite> On the SWE-bench Pro real-world software engineering test, <cite index="16-20">Opus 5 sits just 1.1 points behind the $50/output-token Claude Fable 5 (80.3%).</cite>
On ARC-AGI-3, a benchmark specifically designed to resist memorization and test novel reasoning, <cite index="8-1,8-2,8-3">the ARC Prize Foundation reports a verified 30.16% score at high effort — roughly four times the best previously reported leaderboard score — while GPT-5.6 Sol reached 7.78% and Opus 4.8 reached 1.52%.</cite>
For knowledge work, <cite index="15-2,15-3">Opus 5's 1,861 Elo rating on GDPval-AA v2 leads GPT-5.6 Sol's 1,736 — and even edges out Claude Fable 5's 1,747 — making it the strongest published performer for document-heavy professional tasks.</cite>
On computer use evaluations, <cite index="1-11">Opus 5 surpasses Fable 5's best result on OSWorld 2.0 at just over a third of the cost.</cite>
Safety and Alignment
<cite index="2-13,2-14">Anthropic's automated behavioral audit scores models from 1 to 10 on misaligned behavior, with lower being better; Opus 5 scored 2.30, its best result to date.</cite> <cite index="1-12">Opus 5 also supports zero data retention, unlike Fable 5's 30-day requirement.</cite>
<cite index="5-2">Anthropic said Opus 5 is "the most aligned Opus model and is the least susceptible to being tricked into misuse."</cite> <cite index="5-4,5-5">An Anthropic spokesperson told Axios: "We continue to work with our government partners to conduct their own independent testing of our models. This includes Opus 5."</cite>
Enterprise and Commercial Context
<cite index="19-5">Anthropic's business skews heavily toward API and enterprise usage; according to a February 2026 analysis by Contrary Research, Claude held roughly 40 percent of the enterprise LLM market by usage as of late 2025.</cite> <cite index="19-6">Reuters reported in February that the company was valued at roughly $380 billion in its latest funding round, with annualized revenue climbing from about $1 billion at the end of 2024 to a projected $9 billion by the end of 2025, with internal targets reportedly reaching $20 to $26 billion for 2026.</cite>
Early enterprise adopters reported significant efficiency gains. <cite index="23-3,23-4,23-5">On financial-modeling tasks, Opus 5 averaged 9 percentage points higher accuracy with a third fewer turns and tool calls and 60% less time compared to Opus 4.8.</cite> The platform sits above Sonnet 5 in Anthropic's product hierarchy; <cite index="19-2">Fable 5 is positioned for "the longest, most autonomous jobs, where the model has to stay coherent across many connected steps over hours or days with dense source material."</cite>
<cite index="2-7">A restricted model called Mythos 5 sits above both Fable 5 and Opus 5</cite> and remains available only through invitation-only access programs.