Claude Opus 5 is a large language model from Anthropic that comes close to the performance of the company’s flagship model, Claude Fable 5, at half the price per token. It became available on July 24, 2026, as the fourth Claude model Anthropic released in under two months.
What it is
Claude Opus 5 (API model ID claude-opus-5) succeeds Opus 4.8 in Anthropic’s model lineup, launching through Claude’s chat app, the Claude API, and major cloud platforms — Amazon Bedrock, Google Cloud, and Microsoft Foundry. On Claude.ai, it’s the default model for Max subscribers and the strongest model available on the Pro plan. Anthropic describes it as “a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price,” positioning it as the model for everyday production work — coding, research, and business tasks — rather than pushing the outer edge of what’s possible.
Opus 5 is the fourth model Anthropic released in under two months, after Fable 5 in early June 2026 and Sonnet 5 at the end of that month — part of a broader industry pattern where labs now ship a spread of model tiers at different price points rather than a single do-everything model, so customers can match cost to how demanding a task actually is.
How it compares to Fable 5 and Mythos 5
Claude Fable 5 remains Anthropic’s flagship — the company’s single most capable public model. Opus 5 narrows that gap sharply on several evaluations: it lands within about half a percentage point of Fable 5 on CursorBench 3.2, a test of real coding tasks completed inside a code editor, and it actually beats Fable 5’s best recorded score on OSWorld 2.0 — a benchmark for operating a computer desktop (clicking, typing, navigating apps), the same skill behind computer-use AI agents — while costing about a third as much per task. Anthropic also reports state-of-the-art results among its models on Frontier-Bench and GDPval-AA, an evaluation built around real, economically valuable knowledge-work tasks.
That puts Opus 5 in an unusual position: cheaper than Fable 5, yet close enough in capability that many day-to-day workloads no longer need the pricier model. It’s a distinct model from Claude Mythos 5, Anthropic’s restricted, unfiltered sibling reserved for vetted cybersecurity and biology research partners. Opus 5 is built for general coding, agentic, and enterprise use rather than offensive-security work, and Anthropic says it still trails Mythos 5 specifically on cybersecurity-exploitation tasks.
Safety claims
Anthropic calls Opus 5 its “most aligned model to date,” reporting the lowest rate of deceptive behavior recorded in its automated safety evaluations of any Claude model so far. As with Fable 5, certain biology-research and cybersecurity-exploitation queries are rerouted or restricted rather than answered directly.
What it costs
Through the API, Opus 5 costs $5 per million input tokens and $25 per million output tokens — the same rate as Opus 4.8 before it, and half of Fable 5’s $10/$50 rate, as of July 2026, per Anthropic’s pricing page. A “Fast mode” runs about 2.5 times quicker at double the base price, available through the API and via usage credits in Claude Code. On Claude.ai, it isn’t billed separately: it comes bundled into the Max plan (from $100/month), where it’s the default model, and the Pro plan (from $17–20/month), where it’s the strongest model on offer.
In the news
Anthropic’s July 2026 launch announcement framed Opus 5 as a direct response to the rising cost of running AI-heavy coding and agent workloads at scale — an argument for teams to weigh whether they need Fable 5’s higher price for every task, or whether Opus 5’s near-parity performance already covers most of it.