Nace posts Drex 1.1, and the card titles the weights Drex DLM

Nace posted Drex 1.1 on October 6, 2026, and the card scores Decision Index 0.2 at 52.31 against the kit's Jev cell of 51.67. Weights are CC BY-NC 4.0. Nace's code is MIT. The call is POST /v1/systemone. The hosted Drex 1.5 row on Decision Index 0.2.1 is 58.28.

Nace posted on October 6, 2026, at 17:59 UTC, that Drex 1.1 is now open source. The text calls it an experimental decision model on a diffusion language-model backbone, compatible with /v1/systemone. It names llama.cpp, Ollama, or Unsloth, and the harnesses Claude Code, Codex, Cursor, and OpenCode. The next post is the Hugging Face URL.

The card and the GitHub README are the same text. The title on both is Drex DLM. The files we opened do not print “1.1”. The card says 8 billion parameters, backbone NVIDIA Efficient-DLM-8B, a decision adapter merged in, block tensor names that match Qwen3, and a pointer head. The returns are choice, noul, and score. The route is POST /v1/systemone.

Model context is 32,768 tokens. The local runners default to 16,384. Weights are CC BY-NC 4.0. Nace’s code is MIT. BF16 is about 16 GB. The Q8_0 GGUF is nace-ai/drex-dlm-Q8_0, and the pointer-head matrices stay F16.

The card names three runners. A Python port listens on port 8000. llama-server on nace-ai/llama.cpp, branch edlm, listens on port 8097. An Ollama fork, nace-ai/ollama branch nace-edlm, listens on port 11434. Those three lines do not name Unsloth. The post does. The card says a CUDA build and a CPU-only build were not validated. The tested machine is an Apple M5 Max with 128 GiB. One decision per request is the release contract. The printed ticket is the card’s own local bfloat16 run: billing confidence 0.9461, noul 0.8548, score 1.4078, 87 input tokens. We did not run it.

nace-ai/drex-agent-skill points at this server. It is not a second model.

Decision Index 0.2

The card’s index section is labelled Decision Index 0.2, as of October 2026. It says the other five rows are the top of that board. The kit’s Jev lock on that edition is 51.67.

Model	Index
Drex DLM	52.31
Decider chat Gemma-4-31B	51.93
Jev	51.67
AutoJev-27B	50.94
Jebadiah 27B	50.32
simple-jev, Qwen3.8-27B	50.21

The area cells for Drex DLM are Knowledge and Reasoning 50.71, Language 57.77, Retrieval 54.25, Tools 48.59, and Arts 50.25.

The hosted Drex 1.5 page prints 58.28 on Decision Index 0.2.1, against the September 28 board’s Jev cell of 57.91. That page also prints 199 of 231 against Jev at 201 of 231. The product page already says context moved from 32K on Drex 1.1 to 128K on Drex 1.5, and its FP8 note is about that earlier 1.1.

On October 7, 2026 a search of the Decision Index space for Drex returned no row. The header that day was edition 0.3. The write-up of that board is on the Decision Index page. We did not re-download the September 27 file, and we did not rerun the suite.

This site's reading

Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.

Verify

Primary post is @NaceAI, October 6, 2026, 17:59 UTC, status 2107531163866661105. The text says Drex 1.1 is now open source. It calls the model experimental, names a diffusion language-model backbone, and says the call is compatible with /v1/systemone. The post names llama.cpp, Ollama, or Unsloth, and the harnesses Claude Code, Codex, Cursor, and OpenCode. The follow-up in the same minute, status 2107531343353811145, is the Hugging Face URL.

The card at huggingface.co/nace-ai/drex-dlm and the GitHub README are the same text, opened October 7, 2026. The title on both is Drex DLM. Neither file prints the string 1.1. The card says 8B, backbone NVIDIA Efficient-DLM-8B, a merged decision adapter, block tensor names that match Qwen3, and a pointer head. The returns are choice, noul, and score. The route is POST /v1/systemone. Model context is 32,768 tokens. The local runners default to 16,384. Weights are CC BY-NC 4.0. Nace's code is MIT. BF16 is about 16 GB. The Q8_0 GGUF is nace-ai/drex-dlm-Q8_0, and the pointer-head matrices stay F16.

The card names three runners. A Python port listens on port 8000. llama-server on nace-ai/llama.cpp branch edlm listens on port 8097. An Ollama fork, nace-ai/ollama branch nace-edlm, listens on port 11434. Those three lines do not name Unsloth. The post does. The card says CUDA and a CPU-only build were not validated. The tested machine is an Apple M5 Max with 128 GiB. One decision per request is the release contract. The printed ticket, billing confidence 0.9461, noul 0.8548, score 1.4078, 87 input tokens, is the card's local bfloat16 run. We did not run it.

The card's Decision Index section is labelled Decision Index 0.2, as of October 2026. It says the other five rows are the top of the Decision Index 0.2 board. Drex DLM is 52.31. The area cells are Knowledge and Reasoning 50.71, Language 57.77, Retrieval 54.25, Tools 48.59, and Arts 50.25. Decider chat Gemma-4-31B is 51.93. Jev is 51.67. AutoJev-27B is 50.94. Jebadiah 27B is 50.32. simple-jev, Qwen3.8-27B, is 50.21. That Jev cell is the kit lock, not the 57.91 on the September 28 file.

On October 7 a search of the Decision Index space for Drex returned no row. The space header that day was edition 0.3. We did not re-download the September 27 board file. nace-ai/drex-agent-skill points at this server. It is not a second model.

Compare

52.31 is the card's Decision Index 0.2 cell, against the kit's Jev cell of 51.67. The hosted Drex 1.5 page prints 58.28 on edition 0.2.1, against the September 28 board's Jev cell of 57.91, and 199 of 231 against Jev at 201 of 231. Those figures stay on the Drex 1.5 page. The September 30 product page already says context moved from 32K on Drex 1.1 to 128K on Drex 1.5, and its FP8 note is about that earlier 1.1. This card is the October 6 weight release.

The 0.3 space, opened October 7, ranks Perplexity Decider v1.1 at 62.8 and lists Jev 1.13.0 as a reference at 60.1. A search of that space for Drex found no row. We did not rerun either edition.

Terms

Drex DLM: The card and repository title for the weights Nace posted on October 6, 2026. The post calls the release Drex 1.1. The card does not print 1.1. Backbone NVIDIA Efficient-DLM-8B. POST /v1/systemone. Weights CC BY-NC 4.0. Nace code MIT.
52.31: Drex DLM on the card's Decision Index 0.2 table, against the kit's Jev cell of 51.67. The hosted Drex 1.5 row on edition 0.2.1 is 58.28.