OpenAI

10 days agoNews Editor

OpenAI’s developer account launches COOKED, with Codex building a ramen game live as reset demands flood the comments

OpenAI’s developer account has rolled out a kitchen-themed show called COOKED, where hosts and guests used Codex in its Ultrafast mode to generate an 8-bit ramen game for the new ModRetro handheld. The episode featured AI creator Matthew Berman and editor Brian Knepprath, who described the game by voice while Codex wrote it on the spot. The demo first ran in a desktop emulator and was later loaded onto the handheld, where the game scored the player’s ramen with lines such as 「broth silky」 and 「noodles perfect」. The show also highlighted other OpenAI tools. A persistent agent called dot organized a recipe from photos of ingredients and placed the prep steps into ChatGPT Space, while a GPT-Live-1 voice assistant acted as a cooking copilot at the stove. OpenAI also used the episode to show Codex cloud execution and a refreshed Codex CLI with a fullscreen interface, multi-agent monitoring, and voice input. Still, the reaction quickly shifted away from the demo itself because the episode was framed as day 15 of Codex lead Tibo’s 28-day pledge to ship one improvement every day or reset usage limits for everyone. Commenters repeatedly asked for 「reset」 instead, reflecting frustration over ChatGPT limits that some users said run out within a day and do not feel worth $200.

460

OpenAI

10 days agoNews Editor

OpenAI’s 28-day sprint hits Day 15 with no new feature launch or usage reset

OpenAI’s 28-day product sprint reached Day 15 without a clearly announced new feature or any notice of a usage-limit reset, according to BlockBeats. Instead of unveiling a product update, Tibo shared an episode titled "COOKED" and added the line "We’re cooked." The phrase can mean "we’re done for" in English slang, while also working as a joke in the context of a ramen-cooking segment. The roughly 13-minute video showed OpenAI products while making ramen. In the segment, dot helped prepare a recipe, and GPT-Live-1 guided the cooking process through voice interaction. A developer also used Ultrafast to build an 8-bit ramen game and tested it on a ModRetro handheld. The episode also covered Codex cloud tasks, terminal voice interaction, Agent management, and standalone workspaces. But those features had already been released earlier and were not new Day 15 announcements. Tibo had previously said that each day would bring either a clear improvement or a reset of usage quotas. As of now, neither has been explicitly announced for Day 15.

490

OpenAI

1 months agoNews Editor

GPT-Live-1 tops speech agent ranking after adding Astra, edging past Grok

Artificial Analysis’ latest benchmark shows that OpenAI’s GPT-Live-1 moved to the top of the Speech to Speech Index after connecting GPT-6 Astra as its backend model. The system scored 81.5, slightly above Grok Voice Think Fast 2.0 High at 81.3. According to the benchmark description, GPT-Live-1 handles real-time listening and speaking, while more complex reasoning and tool-use tasks are handed off to Astra. The combined setup did not take first place in speech reasoning alone. Still, it ranked No. 1 in an agent evaluation focused more on practical task execution, posting 67.9%. That result lifted its overall score to the top of the leaderboard. The ranking highlights how model orchestration, rather than a single-model design, can improve end-to-end performance in voice agent testing.

920

OpenAI

1 months agoNews Editor

OpenAI opens ChatGPT real-time voice API at $0.05 per minute

OpenAI has opened access to the GPT-Live-1 API, the real-time voice model currently used by ChatGPT. The model can listen and speak at the same time, understand speech directly, and handle regular conversation without relying on the traditional speech-to-text, large language model, and text-to-speech chain. According to BlockBeats, it can also manage interruptions, pauses, and background noise during a live exchange. For more complex reasoning or tool-use tasks, GPT-Live-1 can hand work off to GPT-6 Astra, Luna, or third-party models, then continue the conversation after receiving the result. Developers are allowed to choose which backend models to connect. OpenAI prices GPT-Live-1 at $0.05 per minute, while backend model usage is billed separately.

1070

Bit.Fan