Cursor launches Composer 2.5: cheaper agentic coding for Australian developers (2026)

What changed

Cursor, the AI coding editor built by Anysphere, has launched Composer 2.5 — its own in-house agentic coding model. It went live today.

Cursor's pitch is blunt. Composer 2.5 matches frontier models on coding benchmarks while costing far less to run. The company points to SWE-Bench Multilingual, a test that measures how well a model fixes real bugs across several programming languages. Cursor says Composer 2.5 reaches parity with leading models there, at roughly a tenth of the token cost.

That cost figure is the headline. "Agentic" coding means the model works across a whole codebase on its own — reading files, editing them, running steps — rather than answering one prompt at a time. That burns a lot of tokens. A model that does the same work for a tenth of the spend changes the maths.

We have not yet seen Cursor's own benchmark write-up confirmed on its blog, so we are treating the parity claim as the vendor's, not ours.

What it means for Australians

For Australian developers and startups, the draw is cost. Agentic coding has been useful but pricey. If Composer 2.5 delivers similar results for less, a small Hobart or Melbourne dev shop can lean on it harder without watching the meter.

The catch: Cursor bills in USD. Pro sits at US$20 a month — about A$28 at our working rate of 1.40, before GST. As an overseas digital service, GST is generally added at checkout. There is no AUD plan and no Australian data residency on offer. For teams handling sensitive client code, that data-handling question still matters.

Does it change our recommendation

Not yet. Our April first look at Cursor 3 already rated it a strong option for AU developers comfortable with USD billing. Composer 2.5 sharpens the cost argument rather than rewriting it.

We will test the new model against real tasks before changing any score. For now, the verdict holds: Cursor remains a serious pick for Australian coders — and Composer 2.5 makes it a cheaper one.

See our earlier take in the Cursor 3 first look. How we score is set out on our methodology page.

Frequently asked questions

What is Cursor Composer 2.5?

Cursor's own in-house agentic coding model, launched 18 May 2026. The pitch: it matches frontier models on coding benchmarks at roughly a tenth of the token cost. We're treating the parity claim as the vendor's until we test it.

Does Cursor charge in Australian dollars?

No. Cursor bills in USD. Pro sits at US$20 a month — about A$28 at our working rate of 1.40, before GST. As an overseas digital service, GST is generally added at checkout. There's no AUD plan and no Australian data residency.

Why does the lower cost matter for Australian developers?

Agentic coding works across a whole codebase on its own, which burns a lot of tokens. A model that does the same work for a tenth of the spend lets a small Hobart or Melbourne dev shop lean on it harder without watching the meter.

Does it change our recommendation?

Not yet. Our April first look already rated Cursor a strong option for AU developers comfortable with USD billing. Composer 2.5 sharpens the cost argument rather than rewriting it — we'll test the model against real tasks before changing any score.

Sources

• Cursor blog — https://cursor.com/blog (Primary, checked 18 May 2026) — primary launch post URL pending confirmation at capture
• TechTimes, "Cursor Composer 2.5 Matches Coding Benchmarks At One-Tenth The Cost" — https://www.techtimes.com/articles/316917/20260520/cursor-composer-25-matches-claude-opus-47-coding-benchmarks-one-tenth-cost.htm (Secondary, 20 May 2026)

Wayback snapshot queued (SPN capture pending).