{"models":[{"name":"qwen3.6:35b-a3b","family":"qwen","generation":"3.6","architecture":"moe","totalParamsB":35,"activeParamsB":3,"sizeGB":23,"license":"Apache-2.0","releasedAt":"2026-04-16","hardwareTarget":"H100","pulledOnH100":true,"defaultInHarness":true,"inRaceRotation":true,"notes":"Harness default since ADR-006. Sparse MoE — only ~3B params fire per token.","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"llama3.2:1b","family":"llama","generation":"3.2","architecture":"dense","totalParamsB":1,"activeParamsB":1,"sizeGB":1.3,"license":"Llama-3.2-license","releasedAt":"2024-09-25","hardwareTarget":"any","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":true,"notes":"Tiniest model in rotation. Hits 100+ tok/s on H100.","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"llama3.1:8b","family":"llama","generation":"3.1","architecture":"dense","totalParamsB":8,"activeParamsB":8,"sizeGB":4.9,"license":"Llama-3.1-license","releasedAt":"2024-07-23","hardwareTarget":"any","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":true,"notes":"Meta baseline. Older than Qwen 2.5; kept for narration/race diversity.","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"qwen3.6:latest","family":"qwen","generation":"3.6","architecture":"moe","totalParamsB":35,"activeParamsB":3,"sizeGB":23,"license":"Apache-2.0","releasedAt":"2026-04-16","hardwareTarget":"H100","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":false,"notes":"Alias for qwen3.6:35b-a3b (same Ollama blob ID).","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"qwen2.5:72b","family":"qwen","generation":"2.5","architecture":"dense","totalParamsB":72,"activeParamsB":72,"sizeGB":47,"license":"Apache-2.0","releasedAt":"2024-09-19","hardwareTarget":"H100","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":true,"notes":"Older dense model. Kept for apples-to-apples comparison with 3.6.","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"qwen2.5:14b","family":"qwen","generation":"2.5","architecture":"dense","totalParamsB":14,"activeParamsB":14,"sizeGB":9,"license":"Apache-2.0","releasedAt":"2024-09-19","hardwareTarget":"any","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":true,"notes":"Mid-size dense. Fits all three Thunder tiers.","updatedAt":"2026-04-17T21:36:15.000Z"},{"name":"qwen2.5:1.5b","family":"qwen","generation":"2.5","architecture":"dense","totalParamsB":1.5,"activeParamsB":1.5,"sizeGB":0.986,"license":"Apache-2.0","releasedAt":"2024-09-19","hardwareTarget":"any","pulledOnH100":true,"defaultInHarness":false,"inRaceRotation":true,"notes":"Small dense. Used on A6000 for baseline race throughput.","updatedAt":"2026-04-17T21:36:15.000Z"}],"count":7}