AI Weekly: Coding Assistants Get Cheaper and Smarter — March 16-22, 2026

The cost of building software with AI just dropped again, and it happened twice in the same week. Cursor shipped a coding model that beats the big labs at a fraction of the price, and OpenAI bought the team behind some of the most-used tools in Python. Neither headline is flashy, but together they tell you where AI coding is heading: cheaper, faster, and more tightly woven into the tools developers already use.

The Deep Dive: Coding Assistants Get Cheaper and Smarter

Cursor's Composer 2 beats Claude on price and benchmarks

On March 19, Cursor released Composer 2, a proprietary coding model built on Moonshot's Kimi K2.5. It scored 61.7 on Terminal-Bench 2.0, ahead of Claude Opus 4.6 at 58.0, and it does that at $0.50 per million input tokens and $2.50 per million output. That is a serious result for a tool most people think of as just a code editor.

The takeaway is not that Cursor "won". It is that a mid-size company can now train a model that outperforms a frontier lab on a specific job, for a lot less money. Specialised beats general when the task is narrow.

OpenAI buys the team behind uv and Ruff

The same week, OpenAI announced it would acquire Astral, the company behind uv, Ruff and ty, three tools a huge chunk of the Python world runs every day. OpenAI folded the team into its Codex effort and mentioned Codex has passed 2 million weekly active users, roughly triple its growth this year.

Buying the plumbing under everyday development is a clear signal. The labs want AI coding assistants sitting right inside the tools you already trust, not off to the side in a chat window.

Why this matters:

• A cheaper coding model that beats the leaders means the price of shipping software keeps falling for small teams.
• When a frontier lab buys core open-source tooling, expect AI to get baked deeper into standard developer workflows.
• OpenAI also released GPT-5.4 nano at $0.20 per million input tokens, making high-volume jobs like classification and support triage genuinely cheap.

What This Means for Australian SMEs

If you pay a developer or an agency to build or maintain software, the economics are shifting in your favour. The tools they use are getting cheaper and better every few weeks, so a project that looked expensive six months ago may be worth a fresh quote now.

If your business runs a lot of repetitive text work, like sorting enquiries or tagging support tickets, the new low-cost models make automating that affordable at last. You no longer need the biggest, priciest model for jobs that are simple but high in volume.

The practical takeaway for this week:

• Ask your developer or agency which coding tools they use and whether a cheaper model could cut your build costs.
• For high-volume text tasks, test a small, cheap model first. You will often get the same result for a tenth of the price.
• Do not chase every new model. Pick one for a real job, measure the result, then decide.

Working out which of these tools is worth adopting, and which is just noise, is exactly what we help clients with at IntelliAgent. If you want a straight answer on where AI could cut your software or admin costs, get in touch.