Claude Opus 4.8 lands: sharper judgement, an effort slider, and a cheaper fast mode

Posted: 9 June 2026 · Category: AI Assistants & Chatbots

Update — 20 September 2026: Anthropic now bills Australians in Australian dollars, GST included. Claude Pro is A$28/month on annual billing (A$340 up front) or A$34/month monthly, and Max starts at A$170.01/month. The pricing page states "Prices include 10% GST". Enterprise seats and API per-token rates stay in US dollars. The USD figures in this piece were correct on the day it was published and are kept as the record. claude.com/pricing — read 19 September 2026.

Anthropic released Claude Opus 4.8 on 28 May 2026. If you use Claude through claude.ai, this is the version you are now talking to. Here is what actually changed, and whether any of it matters if you run a business in Australia.

What changed

Three things stand out.

The assistant got better at judgement. Anthropic says Opus 4.8 is sharper and more reliable on multi-step "agentic" tasks — the kind where you hand it a job and let it work through several steps on its own. The most concrete claim: it is roughly four times less likely than the previous version (Opus 4.7) to let a flaw in its own code slip through without flagging it. That is a coding-specific number, but the broader theme is a model that second-guesses itself more usefully.

There is now an effort control. Next to the model selector on claude.ai (and in Cowork), you can choose how much effort Claude puts into a reply. Turn it up for hard problems and it thinks longer; turn it down for quick questions and it answers faster and uses less of your rate limit. It is available on all plans. In plain terms: a speed-versus-depth dial you set yourself, rather than guessing which model to pick.

Pricing held — and fast mode got cheaper. Standard API pricing is unchanged at US$5 per million input tokens and US$25 per million output tokens. The "fast mode" tier — which runs at up to 2.5× the speed — now costs US$10 input and US$50 output per million tokens, which Anthropic describes as about three times cheaper than the previous fast tier. There is also a coding research preview called "dynamic workflows," which lets Claude plan a large job and run hundreds of parallel sub-agents in one session, aimed at big codebase migrations. That last one is for engineering teams, not everyday users.

What it means for Australians

There was no Australian-specific news in this release. No AUD pricing at the time, no local availability change, no new data-residency commitment — the announcement was standard-global. AUD pricing arrived separately in September 2026; see the note above. So treat it the way you would treat any global Anthropic update.

On price: the API figures are in US dollars. At roughly 1.40 USD to the AUD (as of September 2026), US$5 per million input tokens is about A$7.00, and US$25 output is about A$35.00 — indicative only, before card conversion fees. If you buy through the API as an Australian business, GST treatment depends on how Anthropic invoices you and whether you have provided an ABN; check your tax settings rather than assuming. Most readers, though, use Claude on a flat monthly plan, where this token maths does not apply.

For the typical small-business or professional user, the honest summary is short: the assistant you already pay for got a little sharper, and one usage mode got cheaper. The effort slider is the only change you will actually touch — and it is genuinely handy for trading speed against depth without switching models.

Does it change our recommendation?

No. Claude was already a strong pick in our AI assistants reviews, and Opus 4.8 is an incremental sharpening rather than a reset. Nothing here moves its scores or its standing against ChatGPT and Gemini for Australian users. We will fold the effort control and the updated pricing into our Claude review at its next scheduled update.

If you are already a Claude user, there is nothing to do beyond noticing the new effort dial. If you are choosing between assistants, this release does not change our advice — the model was good before, and it is marginally better now.

Frequently asked questions

What changed in Claude Opus 4.8?

Three things: sharper judgement on multi-step agentic tasks (roughly four times less likely to let a code flaw slip than Opus 4.7), a new effort-control slider, and standard API pricing held while the fast mode got about three times cheaper.

What is the effort slider, and who gets it?

A speed-versus-depth dial next to the model selector on claude.ai and in Cowork. Turn it up for hard problems, down for quick answers. It's available on all plans, and it's the only change most users will actually touch.

Is there any Australian-specific change?

No. The announcement was standard-global — no local availability change and no new data-residency commitment. AUD subscription pricing arrived separately in September 2026. API figures remain in US dollars, and GST treatment depends on how Anthropic invoices you and whether you’ve provided an ABN.

Does it change our recommendation?

No. Claude was already a strong pick, and Opus 4.8 is an incremental sharpening rather than a reset. Nothing here moves its scores or its standing against ChatGPT and Gemini for Australian users.

Sources

• Anthropic, "Introducing Claude Opus 4.8" — https://www.anthropic.com/news/claude-opus-4-8 (checked 9 June 2026)

The AI Directory Australia is independent. We review and score AI tools for Australian users — AUD pricing, AU availability, data residency, GST. We don't write ads; affiliate links are disclosed and sponsored content is labelled. Edited by Adam.