Anthropic's new Opus flagship — near-Fable 5 intelligence at half the price, and a drop-in upgrade over the retired Opus 4.8 at the same rates.
Claude Opus 5, released July 24, 2026, is Anthropic's new Opus-tier flagship and the direct successor to Opus 4.8, which it fully replaces on this site at the same $5/$25 pricing (legacy Opus 4.8 is covered on this page). The pitch is simple: nearly all of Fable 5's intelligence at half the cost. Pricing is unchanged from Opus 4.8 — $5 per 1M input tokens, $25 per 1M output — so anyone on 4.8 gets a step-change in capability for free by swapping the model ID.
Anthropic reports that Opus 5 more than doubles Opus 4.8's score on Frontier-Bench v0.1 at a lower cost per task, scores roughly three times higher than competitors on ARC-AGI 3, and even surpasses Fable 5's best result on the OSWorld 2.0 computer-use benchmark at about a third of the cost. Its biggest gains are on hard agentic coding — multi-file features, large refactors, long autonomous runs — rather than easy single-turn edits.
The headline product feature is the effort dial: five levels from low to max let you trade cost and latency against capability on the same model, and the low settings punch well above their weight. Reasoning is on by default. In the consumer apps, Opus 5 became the default model on Claude Max and the strongest model on Claude Pro at launch; it also powers Claude Code and is available on Amazon Bedrock, Google Cloud, and Microsoft Foundry alongside the Claude API.
Opus 5 is a workhorse for agentic coding and enterprise deliverables — it completes tasks rather than leaving stubs, coordinates teams of sub-agents well, and produces strong Excel models and slide decks. It keeps the high-resolution vision of the 4.8 generation and is notably better at chart, document, and UI understanding. A fast mode (research preview, Claude API only) runs at roughly 2.5× speed for $10/$50 per 1M. Rate limits are a separate bucket from the older Opus 4.x pool.
The honest weaknesses: no image, video, or audio generation — Anthropic stays text-and-analysis only. And like Fable 5, Opus 5 ships with cybersecurity safety classifiers that can decline some security-related requests, though they are notably less restrictive than Fable 5's, and declined requests can automatically fall back to Opus 4.8.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Claude Fable 5 | $10 | $50 | 1M |
| Claude Opus 5 | $5 | $25 | 1M |
| Claude Sonnet 5 | $3 | $15 | 1M |
| Claude Haiku 4.5 | $1 | $5 | 200K |
| GPT-5.5 | $5 | $30 | 1M |
Because Opus 5 launched at identical pricing, it fully replaces Opus 4.8: same rates, strictly better model. Opus 4.8 remains callable on the API for compatibility and serves as the automatic fallback target, but there is no reason to start new work on it. Within the family, Sonnet 5 still gets you most of the way at 60% of the price, and Fable 5 stays on top at 2× the cost for the very hardest long-horizon work. Versus GPT-5.5, Opus 5 matches the 1M context and undercuts it on output.