Images 2.6 puts SVG export, GIFs and markup editing in OpenAI's API
Two new GPT-Image models, Flare and Sunburst, hit the API as the update reaches every tier - plus Kimi K3.1 Pro's open weights and Tencent's Hy4.

Copy markdown
GPT-Image-2.6 brings SVG, GIFs and markup to the API
OpenAI's Images 2.6 is rolling out to every ChatGPT and Codex tier worldwide and ships two API models: Flare, which returns an image in ~6s at low quality, and Sunburst for consistent multi-edit work - generation is ~50% faster overall. The API now does annotation-based markup editing, SVG export and animated-GIF output, so you can script poster layouts, icon sets and short loops. Image tokens run $30/1M output, about $0.025 per 1024² image at low complexity.
Kimi K3.1 Pro's weights are now public
Moonshot put full MXFP4 weights for Kimi K3.1 Pro on Hugging Face (2.8T total, 104B active, with vLLM and SGLang deploy configs); BenchLM ranks it the top open-weight model it tracks (~75). Self-hosting needs a multi-GPU B300 or H200 rack, but the hosted API is $2/$10 per 1M tokens - so you can fine-tune or run a near-frontier agentic model on your own hardware.
Tencent's Hy4 adds computer use via API
Tencent reportedly released Hy4 on Nov 5 with image input and computer-use control through its API at $0.834/$2.501 per 1M tokens, 50% off through Dec 31. Tencent says open weights will follow but no checkpoint is public yet, and the launch benchmarks are all vendor-run - worth a look for agentic UI automation, but verify the numbers yourself.