🌟 Codename: Introducing OpenDesign Arena
🔎 66 PRs · 10 contributors · 7 days — We evaluated over a dozen models using the OpenDesign Harness to compare design quality and cost. DeepSeek V4.1 Flash achieved 98% of top-ranked GPT-6 Astra’s average score at just 1% of its average cost. Explore OpenDesign Arena to find the right model for your design needs and budget.
🔥 Highlights
- 🧭 Runs that keep their evidence. Files, covers, references, produced output, steering, and recovery now travel with a conversation instead of disappearing into a vague finished state. (#7518) Thanks @lefarcen.
- 🛟 Failures that explain what to do next. ACP, model, permission, quota, and runtime problems preserve the useful cause, so a failed run is a decision point—not a dead end. (#6267, #7316, #7429, #7458, #7534, #7647) Thanks @helsome, @Siri-Ray.
- 🎬 Media work that survives the real world. Image and video tasks recover more reliably in packaged Open Design, and Team projects keep their own billing context while they do. (#7504, #7723, #7746) Thanks @alchemistklk, @lbjzz-hash.
- 🪟 The desktop app feels installed, not fragile. First-launch controls, macOS Home readiness, update placement, and app-link handoff now hold up where people actually use the product: the shipped app. (#7479, #7482, #7535, #7688, #7700) Thanks @lefarcen, @Siri-Ray, @alchemistklk.
- 🧩 Previews keep their place. HTML-rich designs, Deck covers, old thumbnail navigation, and edit handoffs retain the structure and state you were working with. (#7421, #7422, #7556, #7558, #7568, #7592, #7651) Thanks @lefarcen, @Siri-Ray.
📥 Download: Tag open-design-v0.22.0.
✨ Added
🧠 Agents, runtimes & sandbox
- 🧰 Codex sessions have a steadier home. Open Design can carry a Codex app-server conversation through streaming replies and tool work without making the result feel like a black box. (#7518) Thanks @lefarcen.
- 🧪 Labs keeps experimentation understandable. Design Harness has a clear settings home, persists its choice, and explains when a policy owns the decision. (#7445, #7514) Thanks @lefarcen.
- 🖼️ Real subjects get real imagery. Ask for a named person, product, artwork, event, or place and Open Design reaches for the real thing instead of a convincing impostor. (#7436)
- 🧑💻 Standalone Terminal work has a real lifecycle. Start, finish, cancel, and return without leaving a ghost session behind. (#7716, #7742) Thanks @PerishCode.
🔑 Models & media
- 🤖 More honest model choices. New model IDs are discoverable, retired models explain themselves, and AMR no longer promises unlimited access where limits apply. (#7401, #7544, #7627, #7686) Thanks @koeseo, @Siri-Ray, @lefarcen.
- 🔗 Team projects remain visible after a bad sync. A transient sync failure no longer hides work your team has already published. (#7781, #7785)
🐛 Fixed
🎨 Studio, editing & canvas
- 🗂️ Open the project you meant to open. Project entry no longer invents facts it cannot know, and artifact-upgrade prompts stay with the active project. (#7428, #7689, #7696) Thanks @lefarcen, @Siri-Ray.
- 💬 A visible question can be answered. Clarification forms keep the run alive long enough to receive the answer—and receive it once. (#7509, #7512)
🖥️ Desktop & setup
- ✅ Onboarding checks the engine before asking you to continue. Invalid runtime setup is caught early, with a path forward instead of a later mystery failure. (#7440, #7743)
- 🌐 The app link returns to the app. Packaged Open Design keeps its normal sign-in and invite handoff while development builds stop claiming the system link. (#7479) Thanks @lefarcen.
🌍 Interface & language
- 📣 The right sunset notice speaks the right language. Go Plan messaging now reaches its intended audience with the selected locale intact. (#7453, #7477) Thanks @nettee.
🙏 Thanks to everyone who shipped 0.22.0
@AmyShang-alt · @PerishCode · @Siri-Ray · @alchemistklk · @helsome · @koeseo · @lbjzz-hash · @lefarcen · @nettee · @simpleqt
