Sign in
Pending benchmarksTracking since Nov 12, 2026

Is Edge0-120B state of the art?Pending benchmarks.Announced Nov 12, 2026; no independent results yet.

Edge0's 120B latent sparse mixture-of-experts model for local coding and agent work, released with Edge0-145B on November 12, 2026. Official model card: 120B total parameters, 8.64B active per token, 48 layers, 512 experts per layer with 10 routed plus 1 shared expert active, and 256K context. Ungated Q3_K and NVFP4 weights are on Hugging Face, with smaller weight files than Edge0-145B. No public hosted API is offered.

Announced facts

The details so far

Lab
Edge0
Announced
Nov 12, 2026
API
Not yet
Open weights
Published · Public, ungated Hugging Face builds Edge0/Edge0-120B-q3k (Q3_K, 27.98 GiB) and Edge0/Edge0-120B-NVFP4 (NVFP4, 32.96 GiB), run locally through edge0-infer and edge0-strata (Edge0 model card and GitHub)
Parameters
120B total, 8.64B active per token
Type
Latent sparse mixture-of-experts language model for local coding and agent tasks, trained natively at 4 bits (Edge0 blog and model card)
Architecture
Official model card: 120B total parameters and 8.64B active per token; 48 hybrid-attention layers (full attention every 4 layers plus a sparse indexer); 512 experts per layer, with 10 routed plus 1 shared expert active per token; expert pool of 60.4B parameters at latent width 1280; 51.2B fp8 n-gram embedding table; 1.26B MTP draft head; 256K context with int8 KV and 32K resident (Edge0 Hugging Face model card)
Base model
Qwen/Qwen3.8-Flash-Next, as stated on the official Hugging Face model cards and the release NOTICE
Inference
Official docs target NVIDIA GPUs with CUDA 13 on Linux or Windows (x64 or ARM64). A local OpenAI- and Anthropic-compatible server is started with edge0-infer serve; the Windows Edge0 Chat app can load a model folder. Apple M2 Ultra Metal figures on the model card are measurements and, the card says, do not establish macOS release availability (Edge0 model card and GitHub)
Hosted access
No public hosted API is offered. Use is local, from the published weights and inference code
License
Edge0 Community Model License 1.0 for Edge0-specific contributions, dated 12 November 2026. Qwen materials remain under the Qwen Community License 1.0; both sets of terms must be followed. The Edge0 license allows personal use, research, modification, noncommercial sharing, internal use, and commercial use of generated outputs, subject to upstream conditions. Selling or commercially bundling the model, operating a commercial model service, or offering an independent commercial coding or office assistant with it requires a separate written Edge0 license (LICENSE, NOTICE, and model card)
Related
Released the same day as Edge0-145B. Separate from Edge0's earlier 35B and 8B models and from its Audio8 speech models

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: self-reported official model card chart scores of 51.3 on DeepSWE 1.1, 75.3 on Terminal-Bench 2.1, 31.0 on HLE, 87.4 on GPQA-Diamond, 79.3 on IFBench, and 81.0 on AA-LCR (Edge0 Hugging Face model card chart).

Lab claim: self-reported Q3_K throughput on an RTX 5090 D 24 GB of 4,576.5 tokens/s for 16K prefill and 144.6 tokens/s agent decode (Edge0 Hugging Face model card).

Receipts

Sources

LABIntroducing Edge0-145B (Edge0 blog)edge0.ai/blog/introducing-edge0-145b-frontier-ai-capabilities-on-your-own-computer · Nov 12, 2026REPOEdge0/Edge0-120B-q3k on Hugging Facehuggingface.co/Edge0/Edge0-120B-q3kREPOEdge0/Edge0-120B-NVFP4 on Hugging Facehuggingface.co/Edge0/Edge0-120B-NVFP4REPOEdge0 Community Model License 1.0huggingface.co/Edge0/Edge0-120B-q3k/blob/main/LICENSEREPOEdge0-AI/Edge0 on GitHubgithub.com/Edge0-AI/Edge0LABEdge0 on X: Edge0-145B and Edge0-120Bx.com/Edge0_AI/status/2120940049391100394 · Nov 12, 2026LABEdge0 on X: weights and inference enginex.com/Edge0_AI/status/2120940898251096122 · Nov 12, 2026LABSamuel Zeng on X: Edge0-145Bx.com/SamuelZengML/status/2120939739431972922 · Nov 12, 2026

Tracking continues until an independent board lists it.