Winnow 12B has 11.96 billion total parameters. The selected Q8_0 GGUF file is 12.67 GB.
Data snapshot: . 2 published tests in this catalog.
Model file and details
Provider
EldanRing
Total parameters
11,959,730,224
Exact file size
12,669,646,592 bytes
Quantization / precision
Q8_0
Weight repository
EldanRing/Winnow-12B
Original model repository
EldanRing/Winnow-12B
Weight revision
fdaebc936c71f831ef5661cbf8d15825f21250e3
Model revision
fdaebc936c71f831ef5661cbf8d15825f21250e3
Runtime notes
Confirm model architecture and quantization support in your runner.
Recorded release date
2026-09-21
Date evidence
Merged Q8_0 weights and the first complete benchmarked release were published 21 September 2026; runtime guidance, NVFP4 and MTP assets were refreshed in commit fdaebc9 on 4 October 2026.
These scores were compiled from the linked sources. These published scores were not independently tested here. Different prompts, precision, thinking budgets and tools can change results.
JevBench public subset · Winnow 231
85.71 / 100
Accuracy on the 231-row public JevBench subset in Winnow's publisher-run candidate-probability harness. This is not the official composite leaderboard score and does not establish unseen-data performance.
Benchmark efficiency: 6.765 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: Publisher evaluation of this exact Q8_0 artifact (SHA-256 b710efc4…18ea) on all 231 rows of the JevBench public subset, using the Winnow llama.cpp decision runtime at 8,192 context, four decision branches, Q8 KV and full GPU offload on an RTX PRO 5000 Blackwell; the publisher recorded 14.36 GiB peak inference VRAM in that run.
Publisher-run result, not an independent test or the official JevBench composite leaderboard score. The public subset may have been visible during base-model pretraining or later development. In a separate publisher RTX 5070 Ti 16 GB profile, Q8 plus the vision projector at 64K context reached 15.01 GiB peak VRAM. Neither measurement is a universal compatibility guarantee: context, KV cache, vision, MTP, runtime and concurrency all affect memory use.
Accuracy on Kev v9's predefined 1,046-row clean/knowable subset in Winnow's publisher-run decision harness. Earlier Kev v4 measurements informed later data refinement, so this is not a wholly untouched task family.
Benchmark efficiency: 6.437 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: Publisher evaluation of this exact Q8_0 artifact (SHA-256 b710efc4…18ea) on Kev v9's predefined 1,046-row clean/knowable subset, using the same pinned 8,192-context Winnow decision runtime.
Publisher-run result, not an independent test. Earlier Kev v4 measurements informed later data refinement; the publisher explicitly says inherited and new-v9 results should not be described as wholly untouched task families.
Context and version evidence: The repository and file name identify the selected model. Revision hashes are shown above when recorded. Tested context length, full hardware configuration and run date are not consistently recorded in this snapshot. Only settings explicitly stated in each benchmark record are known.
The 12.67 GB figure covers model weights in this file. It excludes the context cache, runtime, operating system and any separate multimodal files. It is not a measured RAM requirement or a token-generation speed.
These complete configurations pass the initial weight-size check for the selected 12.67 GB file. Allow extra memory for context and software. Check the model’s runtime notes and your inference software before buying.
Listing evidence: 2026-09-25. Amazon.com base price visible in an owner-supplied screenshot reviewed 25 Sep 2026; capture time not supplied. Dated reference, not a live quote. Excludes card signup offers, coupons, tax and shipping.
Listing evidence: 2026-10-03. Apple US launch price announced 22 September 2026: USD 1,699. Dated manufacturer reference, not an Amazon offer; tax and shipping may differ.
Listing evidence: 2026-09-24. GMKtec, EVO-X2 / 64GB / 1TB, recorded 22 Sep 2026; exact ASIN capacity identified 24 Sep. This is a dated reference, not a current Amazon price; tax, shipping and bundle contents may differ.
Memory matches are estimates, not tested compatibility or speed. As an Amazon Associate I earn from qualifying purchases.
Confirm model architecture and quantization support in your runner. The selected weights use 12.67 GB; this excludes context cache, runtime buffers and other programs.
These are the first listed tiers that pass our reserve policy, not measured minimum requirements. System RAM and VRAM remain separate. Use the memory tool to change the reserve for your context and runner.