← Back to Dashboard

LLM Updates & Breakthroughs

June 2026 — Research-backed analysis of new model architectures and inference breakthroughs.

🧠 VibeThinker 3B: Frontier Reasoning at 3 Billion Parameters

Research

A 3B model matches Claude Opus and DeepSeek V3.2 on math and code. Here's how WeiboAI's Spectrum-to-Signal pipeline makes it possible.

June 29, 2026 · 5 min read

🐦 Ornith 1.0: Self-Scaffolding Coding Agents

New Release

A family of coding models that learn to write their own agent harness during training. The 397B variant beats Claude Opus 4.7 on Terminal-Bench.

June 29, 2026 · 5 min read

⚡ SubQ: First Sub-Quadratic LLM with 12M Token Context

Architecture

A startup claims to have escaped the quadratic attention bottleneck. 12M tokens, 64x less compute, frontier-level retrieval accuracy.

June 29, 2026 · 5 min read

🔬 Qwythos 9B: Claude Mythos Distillation

Fine-tune

A 9B model fine-tuned on 500M+ tokens of Claude Mythos traces gains +34 MMLU and native tool calling. Apache 2.0 licensed.

June 29, 2026 · 4 min read

🚀 DeepSeek DSpark: 60-85% Faster Inference

Inference

Speculative decoding done right: draft model guesses, target model verifies, confidence scheduling saves GPU. Already in production.

June 29, 2026 · 5 min read