← Back to Home

#mixture-of-experts

2 entries with this tag

🔬 research2026-07-15T00:00:00.000Z

Grok 4.5: The Cursor-Trained MoE That Solves SWE-bench Pro Tasks in 4.2× Fewer Tokens

xAI and Cursor jointly release Grok 4.5 on July 8, 2026 — a 1.5T-parameter MoE model trained on trillions of tokens of real developer workflows, hitting 64.7% on SWE-bench Pro, 83.3% on Terminal-Bench 2.1, and 62.0% on DeepSWE 1.0, all at $2/$6 per million tokens with a 500K context window.

#grok#xai#cursor#coding#agentic-ai#mixture-of-experts
🔬 research2026-04-24T00:00:00.000Z

DeepSeek-V4-Pro: Efficient Million-Token Context with Hybrid Attention and MoE Architecture (April 2026)

Analysis of DeepSeek-V4-Pro (1.6T params, 49B activated) and DeepSeek-V4-Flash (284B params, 13B activated) featuring hybrid attention architecture (CSA+HCA), 1M-token context, and three reasoning modes. Comprehensive comparison with frontier models (K2.5, M2.7, GLM-5.1, Qwen3.5-27B, Gemma 4 31B) across reasoning, coding, agentic tasks, and long-context domains.

#ai#models#deepseek#efficient#mixture-of-experts#long-context#reasoning#comparison