Qwen3.8 27B Fast by default, picture to 3D with Idol, and 20-second H3 Turbo videos

Since our October 20 update, giz.ai has a new default model that runs on our own GPU, turns pictures into 3D objects, makes longer videos, and adds voice typing.

Qwen3.8 27B Fast is the new default model

Agents and Auto now start on Qwen3.8 27B Fast, served by giz.ai on our own GPU. It is Swift 1.5, a Qwen3.8 27B tuned to reach its answer without overthinking, with a 262K context, image input, and tool use for coding and agent work.

• Strong on terminal tasks. On Terminal-Bench 2.1, the agentic terminal-coding benchmark, Swift 1.5 scores 72.1% (published by its developer, five runs per task). GPT-6 Luna, our previous default, scored 57% to 60% in our own Terminal-Bench 2.1 run with the Codex agent.
• Much faster. In our tests on the same tasks it matched GPT-6 Luna's scores on reasoning and image questions while answering 2 to 5 times faster: the first words of an Auto reply arrive in about 1.6 seconds instead of 3.6, and prompt enhancement for image and video takes 1 to 2 seconds instead of about 8.
• Twice the daily allowance. Free accounts get 96 calls a day and Plus 960, double GPT-6 Luna's, and other plans scale the same way.

GPT-6 Luna and every other model remain one tap away in the model picker, and characters stay on GPT-6 Luna.

Turn any picture into 3D

Idol (Image to 3D) is new in 3D mode. Give it a picture of an object, or just describe one, and it builds a 3D model you can turn, zoom, and view from every side right in the chat. The background is removed for you, and you can download the result as a PLY file.

Under every generated image there is now a Turn into 3D button that sends the picture straight to Idol. Free accounts can make 3 models a day and Plus 10 an hour.

Longer, sharper videos with H3 Turbo

• H3 Turbo now makes videos of up to 20 seconds, at up to 1280x736, with sound. The longest lengths at the larger sizes use credits.
• Turning an image into a video now starts on H3 Turbo by default.

New image model: FLUX.3 [klein]

FLUX.3 [klein] handles both text-to-image and editing, with up to four reference pictures in one edit. It is now the default for image edits. FLUX.2 Klein on the Plus plan no longer has an hourly limit.

Upgraded models

The included versions moved up to Claude Opus 5.5, Claude Sonnet 5.5, GLM 5.3, Kimi K3, and DeepSeek V4.1 Flash.

Voice typing and read aloud

• Tap the microphone to type by voice: your words appear in the message box as you speak.
• Read aloud lets you choose the speech model, the voice (with search), and the speed. Playback starts within seconds.

Fixes and improvements

• Computer: commands your Agent or a connected AI assistant runs now count as use, so your Computer keeps running while they work. An included Computer stops only after an hour with no use, as the pricing page says.
• Agents: a batch of tool calls is approved once as a whole, reading a web page needs no approval, and a streaming answer keeps going when you switch tabs.
• Downloads: the download button in your generation history works again for every result.

Thank you for using GizAI.