Anthropic says Claude Opus 5 is available today, priced at $5 per million input tokens and $25 per million output tokens, unchanged from Claude Opus 4.8.
That is the whole story in one sentence: a model that Anthropic says “comes close to the frontier intelligence of Claude Fable 5” is being sold at half the price.
And the input rate did not move from the last Opus release. For small teams and solo builders, this changes the default question from “Can we afford the best model?” to “Do we still need to route the hard stuff to Fable 5 at all?”
My take is that Opus 5 will quietly become the workhorse model for coding, agentic tasks, and long-horizon work.
Anthropic’s docs call it a “step-change improvement over Claude Opus 4.8,” with the largest gains in deep reasoning, agentic and long-horizon tasks, and test-time compute scaling.
CNBC reported that Anthropic said Opus 5 outperforms Claude Fable 5 on coding and knowledge work evaluations. And the company said it is “designed to be used every day.” Those are the exact workloads most small agencies and indie devs actually run. So this is a release you route into production, not just bookmark.
Why the price-capability ratio is the real headline
The sticker price is the hook, but the routing math is the lesson.
Opus 5 keeps Opus 4.8’s token rates while jumping to near-Fable performance. That means your existing budget line can suddenly cover a model that Anthropic positions as almost top-tier. If you are running a lean automation shop, that is not incremental news. That is a reason to re-evaluate every model call in your pipeline.
Here is the important caveat Anthropic itself added: CNBC reported that Anthropic said Opus 5 is not the state-of-the-art for “risky, dual-use capabilities,” like cybersecurity. So the model is not a blanket replacement for Fable 5. It is a replacement for the 80% of your workload that is coding, research, writing, and multi-step agent work. The remaining 20%, the stuff where a wrong answer has real downside, still needs the frontier tier or a human review gate.
I am already thinking about this for my own client work. Coding agents, internal tooling, documentation pipelines, and backlog triage are obvious Opus 5 candidates. Anything that touches a live environment, customer data, or adversarial security logic stays on a stricter tier.
The release does not eliminate model hierarchy; it just moves the boundary down to where most of the actual work happens.
Where to find it and what the model IDs mean
Opus 5 is available on Claude API as `claude-opus-5`, through Amazon Bedrock as `anthropic.claude-opus-5`, on Google Cloud as `claude-opus-5`. And in Microsoft Foundry. 9to5Mac reported that Anthropic says it is available today for every platform. That is a full cloud distribution on day one, which matters if your infrastructure is locked into a specific vendor.
On the consumer side, SiliconANGLE reported that Opus 5 is now the default model in the top-end Max tier of the Claude chatbot service. And it is also available in the mid-range Pro plan. MacRumors reported that it is the new default model on Claude Max and the strongest model available on Claude Pro. Fast mode runs at 2.5x the default speed for twice the base price.
If you are paying out of pocket for chat access, that Pro/Max split plus Fast mode is the decision to test first.
For developers, the practical update is a one-line model ID change.
I would still run it in a shadow branch before flipping production traffic. The reason is not the price; the price is stable. Reason: behavior.
The “thoughtful and proactive” warning
Anthropic describes Claude Opus 5 as a “thoughtful and proactive model.” That is marketing language I read with one eyebrow raised. In practice, “proactive” often translates to asking more clarifying questions, surfacing edge cases.
And refusing to finish a task until it is sure it understands the intent.
For an agent that is supposed to run unattended, that can be a feature or a bug.
If the model is more thoughtful, it is too more likely to burn output tokens on explanation.
At $25 per million output tokens, a verbose agent can erase the cost advantage fast.
The same goes for latency.
Anthropic’s gains in test-time compute scaling suggest the model is doing more work per request, which is great for quality and dangerous for timeouts if your pipeline expects a quick turnaround.
So the real cost to watch is not the input rate.
It is the output token count and the end-to-end latency on your longest tasks. My plan is to run Opus 5 against a representative sample of production tasks for 48 hours, compare the output token volume to Opus 4.8. And then decide whether the quality gain is worth the potential speed hit. If you do not measure that, the “half the price” headline will not survive your first monthly bill.
What to do with it this week
If you run a small agency or a solo automation stack, here is the order of operations I would follow:
– Switch coding and knowledge-work pipelines to `claude-opus-5` first. Anthropic says those are the areas where it outperforms Fable 5, so that is where the upside is safest.
– Keep dual-use or high-risk tasks on Fable 5 or a human-in-the-loop workflow. Anthropic explicitly did not claim SOTA on risky capabilities.
– If you are a Claude Pro or Max user, test the default model for a day before paying for Fast mode. 2.5x speed sounds nice, but twice the base price only makes sense when the latency is actually the bottleneck.
– Measure output tokens per request, not just cost per million tokens. A model that writes twice as much can cost more than the one it replaced.
– Update your routing logic so the fallback model is explicit. If Opus 5 starts refusing tasks or over-questioning, your pipeline should degrade cleanly rather than looping.
The broader point is that model tiers are becoming less about raw capability and more about task fit. Opus 5 is the first release that feels like a deliberate pricing move against the “always use the most expensive model” habit.
That is good for lean operators.
The bottom line
Claude Opus 5 is available now, and its release is not just a model upgrade. It is a prompt to redesign your routing table. Anthropic has positioned it as the everyday model that gets close to frontier performance without the frontier price. The question is whether your agents will treat it as a worker or a consultant.
If you route it correctly and measure the bill, it is probably the new default for most of what you build.
Start with one production-adjacent pipeline, run it for 48 hours. And compare the token burn and output quality to your current setup.
If you are already using it, what did you point it at first?
Sources:
– Anthropic announcement: https://www.anthropic.com/news/claude-opus-5
– Anthropic docs: https://platform.claude.com/docs/en/about-claude/models/whats-new-opus-5
– 9to5Mac: https://9to5mac.com/2026/07/24/anthropic-upgrades-claude-with-new-opus-5-model-details-here/
– CNBC: https://www.cnbc.com/2026/07/24/anthropic-claude-opus-5-ai-fable-5-cost.html
– SiliconANGLE: https://siliconangle.com/2026/07/24/anthropic-launches-claude-opus-5-efficiency-safety-improvements/
– MacRumors: https://www.macrumors.com/2026/07/24/anthropic-opus-5/
