496 saved builds published across 18 models

Each published build includes its HTML output, animated previews, a share image, and the original session transcript and runtime/cost metadata. Existing successful new-harness outputs remain in place.

Ministral 8B, Nemotron 3 Nano, Laguna S 2.1 and Qwen3 8B now each have saved new-harness outputs published for all 35 prompts. Other models have partial coverage; failed and unfinished attempts are not included in this release.

• amazon/nova-micro-v1: 34 builds.
• deepseek/deepseek-v3.2: 22 builds.
• deepseek/deepseek-v4-flash: 20 builds.
• deepseek/deepseek-v4-flash-0731: 19 builds.
• google/gemini-2.5-flash-lite: 29 builds.
• mistralai/ministral-3b-2512: 20 builds.
• mistralai/ministral-8b-2512: 35 builds.
• mistralai/mistral-small-3.2-24b-instruct: 32 builds.
• nvidia/nemotron-3-nano-30b-a3b: 35 builds.
• openai/gpt-oss-120b: 33 builds.
• poolside/laguna-s-2.1: 35 builds.
• poolside/laguna-xs-2.1: 20 builds.
• qwen/qwen3-30b-a3b: 33 builds.
• qwen/qwen3-30b-a3b-instruct-2507: 33 builds.
• qwen/qwen3-8b: 35 builds.
• qwen/qwen3-vl-32b-instruct: 13 builds.
• qwen/qwen3-vl-8b-instruct: 20 builds.
• qwen/qwen3.5-9b: 28 builds.