Phai launches Phai-A1 MoE LLM and Phai-M1 video model
11/19/2026 — 11/19, 17:19·1 sources (all)·1 reports (all)
Key takeaways
- Paradigm Intelligence (Phai) quietly launched Phai-A1, a 35B-parameter MoE language model activating about 3B parameters per token, claiming…
Background & analysis
On November 18, Paradigm Intelligence released two models in its Phai series without any prior announcement: the language model Phai-A1 and the video model Phai-M1. Phai-A1 is positioned around speed, with the company stating that at equivalent quality it responds 8–10 times faster than models of a similar parameter scale. Phai-M1 is positioned around realism, emphasizing a weaker sense of artificiality and more refined visual style.
According to test data published by Paradigm Intelligence, under the same single-card RTX 3090 setup and the same version of an Agent task built on the same code, Phai-A1 delivers roughly 8 to 10 times the response speed of Qwen 27B. Public information on ModelScope shows that Phai-A1 uses a MoE (Mixture of Experts) architecture with 350B total parameters, while only around 3B parameters are activated per token. The company explains that Transformer inference compute scales with the number of activated parameters, so reducing the activation count by an order of magnitude also lowers per-token forward computation by an order of magnitude, which it identifies as the main source of the 8–10x speedup. The model is also said to support local deployment on consumer-grade GPUs with 16G of VRAM. No architectural or training details for Phai-M1 have been disclosed by the company.
The release drew substantial attention after launch, and some users reported that the server behind the experience portal crashed on release day. At present, the publicly available information still rests on Paradigm Intelligence's released test figures and the ModelScope listing, with no further technical details on Phai-M1 disclosed so far.
AI-generated from 1 reports · updated 3 days ago
- Heat index
- 25
- All sources
- 1
- All reports
- 1
- First seen
- 3 days ago
Reports on this story headlines open the original
Fanshi released Phai-A1, a 350B-parameter MoE language model that activates only about 3B parameters per token, alongside the Phai-M1 video model. The company says Phai-A1 is 8–10x faster than Qwen 27B on a single RTX 3090 and can run locally on a 16GB consumer GPU; no architecture or training details were disclosed for Phai-M1.
InfoQ 中文AI score 63
Other stories people are talking about
- Heat index 1143RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1011Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
- Heat index 622Anthropic IPO total raise reaches $93.5 billion after greenshoe exercise20 sources
- Heat index 596RisingThinking Machines Lab Open-Sources Inkling 2 Mini with 32GB Model Card Recommendation10 sources
- Heat index 490RisingSEC probes Nvidia accounting practices as shares post seven-week slide9 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

