Composer is a frontier coding model that completes tasks in under 30 seconds.

Introducing Cursor 2.0. Our first coding model and the best way to code with agents.

Easily run the same prompt across models for maximum intelligence.

With a built-in browser, agents can now run and test their code.

Turn speech into code with voice mode.

Read about everything that's new in 2.0. More tomorrow!

For benchmarking purposes, it sounds like Composer falls behind top end models, but excels in speed and likely cost. Would be interesting to test planning with a 'best frontier' model and then executing with Composer while using side by side agents with a Codex or Claude

12 hours of working with Composer on html and css. 12 hours wasted, broke tons, had to restore entire site from backup. Good job guys. 🙂

Loks like @conductor_build

We put Composer through its paces in our vibe check @every

Nice and all, but can you guys get your s*** together and fix because *that* marks you as a most unserious business.

I love this because its so fast, usually when i send a prompt to sonnet or wtvr it takes 2-3 minutes for it to complete which is enough time for me to get stuck scrolling on twitter, not anymore tho!

Nice one Cursor folks, funny to know you adopted the same name we adopted for our major model and platform at @autohand_ai

Pumped to try this out

Prix et efficacité compare à sonnet 4.5 @grok ?

what a stupid layout, normally agent, activity bar, and first side panel (file browser / extensions) is on the left now everything flips in agent mode exactly opposite, and looks like it is even hard coded as neither GUI menu nor 'settings.json' can change that

so far I'm impressed with this model. note: ran into an issue where it tried fixing its own syntax error during a file edit, it tried a couple times then thought to checkout the file and try again. NICE! @ellev3n11 @ericzakariasson amazing work

so i run the same prompt with multiple agents and composer is than reviewing and taking the best approach?

Composer is super fast but it's code error rate is... Something.

Do you run cheetah on groq ?

Good thing about these launches are that there is always something to tinker for the weekend 😄😄

is it made for difficult tasks too ?

Delilik. Aynı ajan birden fazla iş parçacığını paralel yazsa bile (ki bu daha çok işime yarardı) review yapacak olan ben olduğum için, hepsinin hızı benim review hızım kadar şuan. @claudeai 'ın yeni bir sonnet modeli çıkarmasını dört gözle bekliyorum Dario beyden @DarioAmodei 👀

what's this theme

When using the Composer 1 agent with Notion MCP tools, all page update operations fail because the agent incorrectly serializes Python dict parameters as strings instead of JSON objects. This breaks the MCP protocol which expects proper JSON objects. The issue is resolved once switched to Sonnet 4.5.

Incredibly fast AI coding! This is transforming development speed. #AI #CodingTools