Open source · 30 October 2026
Available nowPrime Intellect unveils Prime Mega MoE kernel, 3.11× faster forward pass
Prime Intellect released Prime Mega MoE, a trainable backward pass for DeepSeek’s fused Mega MoE kernel that speeds Qwen3 MoE training steps by up to 34 % in prime‑rl and offers a 3.11× faster forward pass.
Benchmarks are the company's own; not independently verified.
Why it matters
Consider integrating Prime Mega MoE to shorten model‑training cycles; a 34 % speedup can reduce time‑to‑market for new MoE variants.
For product managers
For your product
What does this mean for your product?
Try it on your product More from this issueAll AI news from 30 Oct →
Get this every morning, written for your product.
Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.