APUS Reflex 35B-A3B has 34.661 billion total parameters. The selected Q4_K_M GGUF file is 21.167 GB.
Data snapshot: . 1 published test in this catalog.
Model file and details
Provider
APUS AI-LAB
Total parameters
34,660,610,688
Exact file size
21,166,758,624 bytes
Quantization / precision
Q4_K_M
Weight repository
apus-ailab/APUS-Reflex-35B-A3B-GGUF
Original model repository
apus-ailab/APUS-Reflex-35B-A3B
Weight revision
0103cd6bb0878c4229e5f7ac5f02e20a630d1ffd
Model revision
4947385291375f6f5149b3156ff2811c65ebf020
Runtime notes
Confirm model architecture and quantization support in your runner.
Recorded release date
2026-09-23
Date evidence
APUS AI-LAB created the Apache-2.0 source checkpoint on 22 September 2026 and the same-publisher GGUF repository on 23 September. The exact 21,166,758,624-byte Q4_K_M artifact has a publisher-run full-depth decision-panel result and was rechecked at the 13 October branding revision.
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.
APUS Reflex · Frozen80 Q4
90 / 100
Publisher-reported 80-decision accuracy for the exact full-depth APUS Reflex 35B-A3B Q4_K_M GGUF. Frozen80 is a reused engineering development panel, not a blind general benchmark or an end-to-end browser test.
Benchmark efficiency: 4.252 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-run 72/80 exact-decision accuracy for this Q4_K_M artifact on Frozen80 using llama-server b11118 (commit e6ab7c1a4), an RTX PRO 6000 CUDA system and all 40 model layers. The same prompt tokens were used across BF16 and quantized runs; the Q4 importance matrix came from 448 separate training-course decisions disjoint from Frozen80.
Publisher measurement, not independent validation. Frozen80 is a reused development panel with 80 decisions split evenly across five families, so the score is neither blind nor representative of general chat, generation quality or end-to-end browser success. The Q4 artifact matched the BF16 top decision on 79/80 examples; its reported maximum probability delta was 0.2944. Probabilities are explicitly uncalibrated. The model is text-only, expects 2-16 labeled candidates and emits candidate probabilities; exact probability output uses the publisher client with llama.cpp, while the documented Ollama path requires thinking disabled. The 21.17 GB weight file excludes runtime, KV-cache and context headroom and does not by itself prove compatibility or speed on any computer. Apache-2.0 license.
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 21.167 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 21.17 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: 2026-10-02. Apple US launch price for the base M5 Max Mac Studio, published 22 September 2026: USD 2,499. 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 21.167 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.