GPTImage 2.5 leaks: new image model fakes a GPT-6 launch broadcast convincingly
量子位 · wechat · 2026-09-04
The "host" of the GPT-6 launch was actually faked with the new GPTImage 2.5 image model. Though not officially released, some users are already seeing prompts for it in ChatGPT — the version number jumps straight from 2 to 2.5, suggesting a big upgrade.
Testing of the anonymous luna-lisa-alpha model on LMArena shows much faster generation, more photorealistic output, stable faces across multiple generations of the same character, consistent in-image text, and reduced noise vs GPTImage 2. ChatGPT briefly promoted image generation to the second slot in its main menu, hinting images will be a flagship feature of the GPT-6 era.
Related event: OpenAI's GPT Image 2.5 Leaks Ahead of Launch(4 posts)→
More from Multimodal
- FastH3 Omni open-sourced: reference-conditioned video with synced audio, 5.6-11.9x lower latency — mnmunknown · 2026-10-21
- HeyGen's Professional Voice Clone needs just 30 min of audio to sound exactly like you — rohanpaul_ai · 2026-10-21
- Flux 3 Max ranks #1 on physics with 1.5M+ human judgments — isidentical · 2026-10-21
- Claude Code directed a full music video: 6.5 hours, 765M input tokens — colinmcnamara · 2026-10-21
- Kling 4.0 debuts native 30-second AI video with 10 keyframes, extendable to 2 minutes — lmoroney · 2026-10-21
- Mistral's Voxtral transcribes an 1891 Eiffel phonograph recording at 6.5% WER — IsForAt · 2026-10-21