Meta ships a $0.10/$0.20 Muse Spark tier — the catch is your data
A 1M-context multimodal model at a 12x discount — if your prompts train Meta. Plus Cursor's always-on agents and transparent backgrounds in gpt-image-2.
Copy markdown
A 12x price cut, paid in training data
Muse Spark 1.2's new Contributor tier runs $0.10 in / $0.20 out per million tokens — vs $1.25/$4.25 for the standard tier — on the same 1M-context model that takes text, images, video, audio, and PDFs. The trade: your prompts and outputs may be used to train Meta's models, so it's for prototyping and learning, not client work.
Cursor agents you don't have to babysit
Cursor's new Subscriptions let an agent watch your PRs, follow a Slack thread, or run on a schedule; custom modes pin skills as 'always on,' subagents get isolated VMs, and `/goal` holds a long-lived objective across runs. It's the shift from one-shot prompts to standing background agents.
Transparent PNGs straight from gpt-image-2
OpenAI's image model can now return a transparent background — set `background` to `transparent` with `png` or `webp` output. That's drop-in logos, icons, and game sprites with no separate background-removal pass.
Copilot's coding agent lands in Slack
After Microsoft Teams, GitHub's Copilot coding agent now works inside Slack — assign it an issue, review its plan, and merge from a thread. Handy if your team lives in Slack channels rather than the GitHub web UI.