Pending benchmarksTracking since Nov 4, 2026
Is IOTA Decide 4B state of the art?Pending benchmarks.Announced Nov 4, 2026; no independent results yet.
Macrocosmos's first public model: a 4.2B decision model fine-tuned from Qwen3.5-4B on GPUs spread over the internet, returning calibrated probabilities rather than generated text.
Announced facts
The details so far
- Lab
- Macrocosmos
- Announced
- Nov 4, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- 4.21B (4,205,751,296)
- Base model
- Fine-tune of Qwen/Qwen3.5-4B
- Version
- v0.4 (earlier v0.3 and unversioned repos exist on Hugging Face)
- Output
- Calibrated probabilities over decisions; no free-text generation
- Training
- 16 RTX A4000 GPUs connected over the internet (lab-stated)
- License
- Apache-2.0
Tags marked unsure are not stated on the lab's own pages.
Next steps
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
Self-reported
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: 78.6 composite on JevBench, measured by Benchmark Heaven on Nov 4, 2026 (board listing pending).
Lab claim: about 33 ms per decision.
Receipts
Sources
LABMacrocosmos on X: IOTA Decide 4B, our first public modelx.com/MacrocosmosAI/status/2118028289813971098 · Nov 4, 2026REPOmacrocosm-os/iota-decide-4b-v0.4 on Hugging Facehuggingface.co/macrocosm-os/iota-decide-4b-v0.4Tracking continues until an independent board lists it.