Claude Opus 5 Is Officially Here — 1M Context, a New 'xhigh' Mode, and the Rumors Were Right

The rumor mill called this one correctly. Two weeks after we covered the 'Honeycomb' leak and speculation around a Claude Opus 5 launch, Anthropic made it official on July 24: Claude Opus 5 is live, and nearly every leaked detail — the codename sighting, the context window, even the exotic reasoning mode — checked out.

The headline spec is a 1-million-token context window with up to 128,000 tokens of output in the standard API (300,000 via the Batch API). Opus 5 also ships a new reasoning-effort level called "xhigh," sitting above today's "high" and below a new uncapped "max" tier — built for long-horizon agentic and coding work that can run 30+ minutes and burn through multi-million-token budgets. Notably, on Opus 5, thinking can't be switched off at xhigh or max; Anthropic's API will reject the request rather than let you run a shallow pass at that effort level.

Positioning matters here: Opus 5 does not replace Fable 5 as Anthropic's top-tier frontier model — that crown stays with Fable 5 (and its invite-only sibling, Mythos 5). Instead, Opus 5 replaces Opus 4.8 as the workhorse tier, inheriting its exact pricing: $5 per million input tokens, $25 per million output tokens, with a "Fast mode" running roughly 2.5x quicker at double the base price. Opus 4.8 is now a legacy model, and Opus 4.1 is being deprecated outright, retiring August 5 with Anthropic pointing existing users toward Opus 5.

On benchmarks, Anthropic leans on its own eval suite — Frontier-Bench, CursorBench, ARC-AGI 3, an internal automation benchmark, and coding-specific tests — rather than headline industry leaderboards. The one figure worth flagging: Anthropic says Opus 5 at max effort lands within half a percent of Fable 5 on CursorBench, at roughly half the cost — a pitch aimed squarely at teams who want near-flagship coding performance without paying flagship prices.

The timing lines up with everything the rumor pointed at. Kimi K3 topped a coding leaderboard ahead of Fable 5 in mid-July at a noticeably lower price, and Fable 5's free access window closed July 19 right as premium-plan demand was reportedly straining capacity. A refreshed, identically-priced Opus tier with near-flagship coding scores is a direct answer to both pressures at once — reasserting a value floor under Fable 5 without actually discounting it. Whether "xhigh" earns its keep in production agentic workloads, or turns into another effort knob most people leave on default, is the next thing worth watching.

Follow-up analysis: we've since put those numbers side by side two ways — Opus 5 against the rest of Anthropic's lineup, where it beats the double-priced Fable 5 on the coding charts Anthropic itself published, and Opus 5 against GPT-5.6, where that "knob most people leave on default" turns out to match GPT-5.6 Sol at maximum effort.