- Parameters
- 180B / 6B active
- Quant
- AD-3.84bpw M64
- Size
- 79.1 GB
- Family
- qwen3.x
qwen38-flash-next
RemovedMoEdeletedllama.cpp
Model card ↗Outcome as of 20 September 2026, from the seat registry. The write-up below was made when the model was tested.
Why it was tested
Tested because the M64 shard layout claims to run a 180B model on 64GB by keeping a 51B n-gram embedding table on SSD.
What came of it
Deleted 2026-09-09 (88 GB). Lost to REAP-60b on every suite where both were measured: screening 95.0 vs 98.3, grounding 81.6 vs 87.2, nullschema 58.3 vs 97.2, chainwalk 0/2 (reached steps 5 and 13 of 200). nullschema is abstention, so 58.3 means it fabricates on ~40% of absent-value cases - the worst weakness for accountancy extraction. It answered the question it was downloaded for (does 40s expert streaming buy better task completion?) and the answer was no. Restore notes in ~/models/RESTORE-deleted-20260909.sh. One reason to revisit: oMLX ships mlx_vlm_qwen4_exp_compat, so it may have an MLX path mlx_lm cannot offer.
| Suite | Score | 95% interval | Passed | Runtime · profile | Recorded |
|---|---|---|---|---|---|
| grounding | 79.2% | 71.3–85.4 | Passed 99/125 | llama.cppextraction | 2026-09-05 |
| nullschema | 50.0% | 38.7–61.3 | Passed 36/72 | llama.cppextraction | 2026-09-05 |
| screening | 96.7% | 88.6–99.1 | Passed 58/60 | llama.cppextraction | 2026-09-05 |
Runs marked unenforced were measured through mlx_lm on a suite that requires constrained decoding. mlx_lm accepts the schema and ignores it, so those numbers do not measure what the suite is designed to measure. They are kept and labelled rather than deleted.
Every run log entry for this model (3)
| Finished | Suite | Runtime · profile | Score | Coverage | Grader · dataset · config |
|---|---|---|---|---|---|
| nullschemanullschema/1 | llama.cppextraction | 50.0% | Coverage 72/72 | g d7448e0f5fc3f9afd 3b91280bb40372efc d248e89da830ce31 | |
| screeningscreening/1 | llama.cppextraction | 96.7% | Coverage 60/60 | g 0661b49a10be2d2ad 64c97ca2269c29afc dd9b4d06d5abee09 | |
| groundinggrounding/1 | llama.cppextraction | 79.2% | Coverage 125/125 | g 30b933dcc6a7e2d4d dcb396faf8b83b0ac 9148fcdc2bc14fe4 |
- Qwen3 Coder 30BMeasured, not seated
30B · MoE · Q4_K_M · 18.0 GB
The first coding model measured against a real coding contract rather than an extraction suite.
- Qwen3.5 122B-A10BRemoved
122B · MoE · IQ3 · 45.0 GB
Tested as the largest MoE that would fit, on the assumption that more parameters would win.
- Qwen3.6 35B-A3BMeasured, not seated
35B · MoE · MLX 4-bit (also held as UD-Q5_K_M GGUF) · 19.0 GB
Tested as the fast general MoE, and used as the daily coding driver in DSH. Moved to an MLX build on 2026-09-09.
- Qwen3.8 27BSeated · agentic
27B · dense · MLX 4-bit (MTP build, template patched to reasoning_effort=medium) · 15.0 GB
The dense counterweight to the MoE fleet, and the model where MLX first beat llama.cpp. Promoted to the agent seat on 2026-09-09 after toolcall/1.