What Is Claude Opus 5? Features, Pricing, and Uses

Claude Opus 5 is the model most Claude users will now reach for by default, and the reason is pricing rather than a raw capability jump. Anthropic shipped it at the exact rate of the model it replaced while claiming near-flagship performance, which reframes the usual capability-versus-budget tradeoff. This guide explains what Claude Opus 5 is, what changed, how it performs, what it costs, and where it fits against the rest of the Claude lineup. For the deep data behind each of those, it points you to the focused breakdowns rather than repeating them here.

What is Claude Opus 5?

Claude Opus 5 is Anthropic's Opus-tier model in the Claude 5 family, positioned as a near-frontier system that comes close to the more expensive Fable 5 at half the token price. Anthropic released it on July 24, 2026, using the API name claude-opus-5, and calls it a thoughtful and proactive model built for everyday use rather than a specialist ceiling.

The lineup context matters. Above Opus 5 sit Claude Fable 5 and the gated Mythos tier. Below it sits Sonnet 5, the volume workhorse. Opus 5 is aimed at agentic coding, computer use, and long-horizon reasoning, the work where a wrong answer is expensive but the top tier is overkill. Anthropic developed it under the pre-launch codename Honeycomb.

Its headline specs are a 1 million token context window as both default and maximum, and a standard maximum output of 128,000 tokens. That context size lets Opus 5 work across large codebases, long documents, and multi-step agent runs without losing earlier context.

What's new in Claude Opus 5

1. Effort dial and thinking by default

The defining change in Claude Opus 5 is an effort dial paired with thinking that now runs by default. Developers can set a low, medium, or high effort level per request, raising reasoning compute for hard problems or lowering it to conserve tokens on routine work. Because extended thinking is on by default, the same request emits more tokens than it did on Opus 4.8, so an identical price per token can still produce a larger invoice.

2. Two dropped features

Two features from Opus 4.8 do not carry over, and both matter before you migrate. The migration specifics, including the exact benchmark deltas and one breaking API change, live in our Opus 5 vs Opus 4.8 breakdown.

3. Fast mode and beta updates

Alongside the model, Anthropic shipped Fast mode, which runs Opus 5 at roughly 2.5 times default speed for twice the base price. It also released two beta updates: mid-conversation tool changes and automatic API fallbacks, which route requests flagged by safety classifiers to another model rather than blocking them.

How Claude Opus 5 performs

1. Overall benchmark standing

Claude Opus 5 leads on agentic and knowledge-work benchmarks while sitting a hair behind Fable 5 on a few coding evaluations, and the pattern is consistent across independent and vendor sources. On the independent Artificial Analysis Intelligence Index, a composite of nine evaluations, Opus 5 ranks first with a score of 63 at max effort (snapshot August 2026; the score is effort-dependent and updates with the Index version, so reconfirm at publish). Anthropic reports it as state-of-the-art on Frontier-Bench v0.1 and GDPval-AA, where it more than doubles Opus 4.8's result at a lower cost per task.

2. Standout results

The single most-cited result is ARC-AGI-3, an evaluation of how well a model solves novel problems it has not seen. Anthropic reports Opus 5 scoring roughly three times the next-best model, a jump large enough that it reads as a capability arriving rather than one improving. On computer use, Anthropic says Opus 5 surpasses Fable 5's best OSWorld 2.0 result at just over a third of the cost.

3. Where it falls behind

One boundary is worth stating plainly: Opus 5 is not the cybersecurity leader. Anthropic's own disclosure places it behind the Mythos tier on offensive cyber tasks, the one category where it does not claim the top Claude spot.

Where the numbers get granular, per harness and per source, the reviews and comparison spokes carry the full tables. For what testers and real users make of it beyond the benchmarks, our Opus 5 reviews roundup covers the love-hate split in detail.

What Claude Opus 5 costs

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, identical to Opus 4.8 and exactly half of Fable 5. That flat pricing is the whole story of the launch: you are not trading capability against budget when you move from 4.8, because every line of the rate card is unchanged.

The rate card is the easy part. What actually decides your bill is the effort setting, how much of your context repeats, whether thinking tokens run by default, and a short list of modifiers. Fast mode doubles both sides to $10 and $50; the Batch API halves them. The full arithmetic, including which charges never show up in a per-token table, sits in our Opus 5 pricing breakdown.

How Claude Opus 5 compares

Claude Opus 5 makes the most sense when you see it against its neighbors, and each of those matchups has its own detailed scorecard. Rather than compress them here, this section points you to the right one for your decision.

1. Within the Claude line

Within the Claude line, the live question is whether Fable 5 still earns its premium now that Opus 5 sits at half the price. Our Opus 5 vs Fable 5 breakdown separates what Anthropic reported from what independent evaluators measured and ends with a routing call by workload.

2. Against other vendors

Against other vendors, two comparisons matter most. Gemini 3.7 Flash vs Opus 5 runs both through the same nine-benchmark suite and lands on a workload split rather than a single winner. For the open-weight angle, GLM 5.3 vs Opus 5 covers the coding-versus-reasoning divide and the self-hosting question Opus 5 cannot answer.

3. If you are switching away

If you are weighing whether to switch away entirely, our Opus 5 alternatives guide maps the wider field, from GPT-5.6 Sol to the open-weight contenders. For the tier above, our Fable 5 alternatives roundup covers the same question at the frontier end.

Who should use Claude Opus 5

1. When Opus 5 is the right pick

Claude Opus 5 fits teams shipping products in large codebases, running multi-step agents, or doing demanding knowledge work where accuracy pays for itself. Anthropic positions it as the everyday model, and the benchmark pattern backs that: it owns the daily-agent cluster of terminal coding, knowledge work, novel problems, search, and desktop control.

2. When to reach for another tier

Reach for a higher tier only when your work sits at the specialist peaks. Mythos-tier science and offensive cybersecurity still call for the top of the range, and the longest autonomous agent horizons may justify Fable 5. Step down to Sonnet 5 when most of your work is routine and high-volume, where its cost curve wins. For most production work between those extremes, Opus 5 is now the default choice, and its identical pricing to Opus 4.8 removes the usual reason to hesitate.

For conversational and orchestration-heavy work, weigh the tradeoffs testers raise before committing, since capability leadership and day-to-day pleasantness are not the same thing here.

Build with Claude Opus 5 on Emergent

Claude Opus 5 is now available on Emergent, the AI app-building platform that turns a plain-language description into a working, deployable, full-stack application. You do not need to write code or manage an Anthropic account to build with it. Emergent handles the backend, database, auth, payments, and deployment, and you own the code that comes out.

Model access runs through the Universal LLM Key, a single credential that also covers GPT from OpenAI and Gemini from Google, all billed through Emergent Credits with no separate API setup. You select the model per project, so you can put Opus 5 behind the work that needs it and choose differently for the next build. Describe the app you want and let Opus 5 do the building.

Start Building with Emergent.