Published
Updated
Opened October 26, Decision Index 0.4.1 Full score for d3-turbo is 67.31, and vision full is 72.46
Xunzhuo Liu posted d3-turbo on October 20, 2026, at 00:42 UTC. Opened October 26, Decision Index 0.4.1 Full score is 67.31 and vision full is 72.46, second of 56. The card's own 0.3.2 estimate is 70.2. Apache-2.0. This desk did not run the weights.
Xunzhuo Liu posted d3-turbo on October 20, 2026, at 00:42 UTC. The attachment is a video of about 41 seconds. The post names a multimodal foundation decision model, Box and track, StateOnce about 15 times faster than d3-flash, and links vllm-sr/d3-turbo. A follow-up at 00:49 UTC prints 255 questions about one photo at 2.12 s against d3-flash at 32.94 s. At 00:55 UTC he wrote that the free API at decision.vllm-sr.ai serves d3-turbo, and that vllm-sr/d3 names the same model. The share image on this page is a Box example still from the card. The earlier Decision 3.0 family is on Decision 3.0. Sibling d3-ultra has its own page.
The card, opened October 27, prints 26.09B parameters including a 0.46B vision encoder, Choice, Yes / No, Score, Box, and Track, and 131,072 tokens per question with long_context=True (32,768 by default). License is Apache-2.0. The highlights print a lab estimate of 70.2 on Decision Index 0.3.2 text against d3’s official board cell of 66.3, and 72.9 on the vision board against d3’s 71.3. The same card prints a public-suite run with the official 0.3 kit at 71.7 against d3’s 65.5. Box accuracy at IoU 0.5 is 0.90, and confidence AUROC is 0.96. Track at two updates per second prints success AUC 0.80 and presence F1 0.97. On one AMD Instinct MI325X, 255 questions about one photo take 2.12 s for d3-turbo and 32.94 s for d3-flash. The collection is Decision 3.0. The repository named on the family cards is vllm-project/semantic-router.
Opened October 26, 2026, 15:30 UTC, data/index.json is Decision Index 0.4.1, 229 models, generated 2026-10-26T15:30:58Z. Full score for vLLM-SR d3-turbo v1.1 is 67.31, eleventh of 229. Public skill is 71.62. d3 on that file is 67.15. d3-ultra is 70.18. Jev 1.14 is a reference at 60.24. The vision board data/vision.json, generated 2026-10-26T15:31:01Z, edition 0.4, 56 entrants, prints d3-turbo full 72.46, public 76.84, private 68.08, rank 2. RBDash-Decision v0.2 is first at 73.3. d3 on that vision file is 71.26. We did not rerun either board.
Tinfoil, October 26, 15:16 UTC, says its decision-large, decision-medium, and decision-small sizes are powered by vLLM’s d3 family. This desk did not call Tinfoil.
Date header 06:25:57 GMT October 27, GET https://decision.vllm-sr.ai/v1/models returned HTTP 200 and listed vllm-sr/d3-turbo, with vllm-sr/d3 as an alias of that id. Date header 06:25:58 GMT, POST /v1/systemone without a key returned HTTP 401, missing_api_key. This desk has no key for that host. We did not run the weights.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Primary post is @XunzhuoLiu, October 20, 2026, 00:42 UTC, status 2112343603209261092. The text names d3-turbo, multimodal foundation decision model, number one on both Jev Decision Index boards in the lab's reading, Box and track, StateOnce about 15 times faster than d3-flash, and links huggingface.co/vllm-sr/d3-turbo. The attachment is a video of about 41 seconds. Status 2112345370789621909, 00:49 UTC, prints 255 questions about one photo at 2.12 s against d3-flash at 32.94 s, 15.5 times. Status 2112346906345210071, 00:55 UTC, says the free API at decision.vllm-sr.ai serves d3-turbo, and that vllm-sr/d3 names the same model. The share image on this page is a Box example still from the card.
The Hugging Face card vllm-sr/d3-turbo, opened October 27. Parameters 26.09B including a 0.46B vision encoder. Decision types Choice, Yes / No, Score, Box, and Track. Context 131,072 tokens per question with long_context=True, 32,768 by default. License Apache-2.0. Highlights print a lab estimate of 70.2 on Jev Decision Index 0.3.2 text against d3's official 66.3, and 72.9 on the vision board against d3's 71.3. Public suite with the official 0.3 kit: 71.7 against d3's 65.5. Box accuracy at IoU 0.5 is 0.90, confidence AUROC 0.96. Track at 2 updates per second prints success AUC 0.80 and presence F1 0.97. Serving table on one AMD Instinct MI325X: 255 questions 2.12 s for d3-turbo against 32.94 s for d3-flash. Free API decision.vllm-sr.ai. Collection Decision 3.0. Sibling d3-ultra is on its own page. The earlier family page is Decision 3.0.
Opened October 26, 2026, 15:30 UTC, the Decision Index space data/index.json is edition 0.4.1, 229 models, generated 2026-10-26T15:30:58Z. Full score for vLLM-SR d3-turbo v1.1 is 67.31, eleventh of 229. Public skill is 71.62. d3 on that file is 67.15. d3-ultra is 70.18. Jev 1.14 is a reference at 60.24. The vision board data/vision.json, generated 2026-10-26T15:31:01Z, edition 0.4, 56 entrants, prints d3-turbo full 72.46, public 76.84, private 68.08, rank 2. RBDash-Decision v0.2 is first at 73.3. d3 on that vision file is 71.26. We did not rerun either board.
@TinfoilAI, October 26, 15:16 UTC, status 2114737835177865675, says decision-large, decision-medium, and decision-small on Tinfoil are powered by vLLM's d3 family. This desk did not call Tinfoil.
Date header 06:25:57 GMT October 27, GET https://decision.vllm-sr.ai/v1/models HTTP 200. The list includes vllm-sr/d3-turbo, and vllm-sr/d3 as an alias of that id. Date header 06:25:58 GMT, POST /v1/systemone without a key returned HTTP 401, missing_api_key. No DECISION_API_KEY in this desk's environment. We did not run the weights.
Compare
67.31 is Full score on Decision Index 0.4.1. 70.2 is the card's own estimate on edition 0.3.2. 72.46 is vision full on edition 0.4. 71.7 is the card's public-suite print with the 0.3 kit. Do not merge those four. Do not merge 67.31 with d3-ultra's 70.18 on the same 0.4.1 file, or with d3's card cell 66.3.
2.12 s is 255 questions for d3-turbo on the card's MI325X table. 32.94 s is d3-flash on that same table. One desk POST without a key is not a latency ranking against those clocks.
Terms
- d3-turbo
- vLLM Semantic Router's upgraded 26.09B Decision 3.0 checkpoint, Apache-2.0. Opened October 26, 2026, Decision Index 0.4.1 Full score is 67.31. Vision full on edition 0.4 is 72.46. The free API aliases vllm-sr/d3 to this model.
- State once
- The card's name for reading the input once per request and answering every question from that read. On the printed MI325X table, 255 questions about one photo take 2.12 s for d3-turbo against 32.94 s for d3-flash.