| Claude Opus 5.5 | 0.268 | score | Five selected episodes; earlier guide d69e7f1e and stricter text-only GPT-6 Astra judge; 104 fixed segments, 145 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; Opus via OpenRouter; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-10-29 · document_publication | Shotwell-WGO-Bench · original five-episode pilot results |
| GPT-6 Astra | 0.442 | score | Five selected episodes; earlier guide d69e7f1e and stricter text-only GPT-6 Astra judge; 104 fixed segments, 145 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-10-29 · document_publication | Shotwell-WGO-Bench · original five-episode pilot results |
| GPT-6 Luna | 0.178 | score | Five selected episodes; earlier guide d69e7f1e and stricter text-only GPT-6 Astra judge; 104 fixed segments, 145 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-10-29 · document_publication | Shotwell-WGO-Bench · original five-episode pilot results |
| GPT-6.1 Sol | 0.423 | score | Five selected episodes; earlier guide d69e7f1e and stricter text-only GPT-6 Astra judge; 104 fixed segments, 145 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-10-29 · document_publication | Shotwell-WGO-Bench · original five-episode pilot results |
| Gemini 4 Argon | 0.355 | score | Five selected episodes; earlier guide d69e7f1e and stricter text-only GPT-6 Astra judge; 104 fixed segments, 145 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; two refused windows count as missing annotations; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-10-29 · document_publication | Shotwell-WGO-Bench · original five-episode pilot results |
| Claude Opus 5.5 | 0.323 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; Anthropic API; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| Claude Sonnet 5.5 | 0.254 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; Anthropic API; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| GPT-6 Astra | 0.548 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| GPT-6 Luna | 0.204 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| GPT-6.1 Sol | 0.511 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| GPT-6.1 Terra | 0.494 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| Gemini 4 Argon | 0.466 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| Kimi K3 (1,024 reasoning tokens) | 0.162 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; reasoning capped at 1,024 tokens; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| Kimi K3.1 Pro (1,024 reasoning tokens) | 0.217 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; 8-fps 512-pixel frames in 4x2 contact sheets; 10-second scoring cores with 1-second context; reasoning capped at 1,024 tokens; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |
| NVIDIA FLORA | 0.354 | score | Eleven selected Job 690 episodes; guide 6bbcf11f7d9c and reference 5ff9337869cb; updated core-action-v2 text-only GPT-6 Astra judge; 221 fixed segments, 325 required details; NVIDIA native TempSeg + Captioner pinned 78a0e4ac; source video 8 fps, internal temporal sampling 4 fps; condensed <=2,000-character instruction; printed Overall F1 is segmentation F1 × description F1; reported_at is author source revision date, not an inferred experiment date. | 2026-11-02 · document_publication | Shotwell-Bench · eleven-episode results at author commit 79d00b3 |