Four frontier models face the same 3D prompt as GPT-6 Astra, Claude Fable 5.1, Kimi K3 and GPT-5.6 Sol stack up

OpenAI’s newly previewed GPT-6 Astra, Anthropic’s Claude Fable 5.1, Moonshot AI’s Kimi K3 and OpenAI’s GPT-5.6 Sol were given the same 3D-scene prompt, and the outputs diverged sharply—underscoring how quickly the frontier pack is splitting on visual generation, cost and reliability. Developers who ran the side-by-side said Astra produced a clear jump over Sol on 3D scenes, while Kimi K3 delivered competitive results at a fraction of the spend; Fable 5.1, by contrast, came back looking lazy on that particular task and still looked odd after two regenerations.