← Catalog
Phai-A1-35B-1M
AvailablePhancy GroupOpen source
Phancy Group's coding and agent language model, launched Nov 18 2026 with the Phai-M1 video model. Built on Qwen3.6-35B-A3B (MoE, about 35B total / 3B active) and post-trained (SFT plus GRPO reinforcement learning across 4,000 sandbox environments) entirely on Huawei Ascend NPUs. Supports a 1M-token context via YaRN (factor 4 over 262,144). Weights are on Hugging Face (PhaiModelLab/Phai-A1-35B-1M) and ModelScope under Apache 2.0 and can run locally. Phancy reports SWE-bench Verified 83.6% and SWE-bench Pro 65.0% (vendor-reported, unverified).
Specifications
- License
- Open source · Apache-2.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 35B · 3B active
- Context window
- 1.0M tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- TextCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.