200B Model Beats DeepSeek-R1 – ByteDance Launches Seed-Thinking-v1.5

ByteDance's Seed-Thinking-v1.5: 200B MoE model beats DeepSeek-R1 in reasoning & STEM tasks. Outperforms in benchmarks like AIME & Codeforces.

"AI Disruption" Publication 5800 Subscriptions 20% Discount Offer Link.

---

The ByteDance Doubao team today released a technical report for their new inference model Seed-Thinking-v1.5.

According to the report, this is a Mixture-of-Experts (MoE) model with a total of 200 billion parameters, activating 20 billion parameters during operation.

Its performance is remarkably impressive, surpassing DeepSeek-R1, which has 671 billion total parameters across various benchmarks.

Some speculate that this is the deep reasoning model currently in use by ByteDance’s Doubao.