// Latest Model News
🧠 VibeThinker 3B: Frontier Reasoning at 3 Billion Parameters
ResearchA 3B model matches Claude Opus and DeepSeek V3.2 on math and code. We break down the Spectrum-to-Signal pipeline.
July 1, 2026 · 5 min read
🐦 Ornith 1.0: Self-Scaffolding Coding Agents
New ReleaseA family of coding models that learn to write their own agent harness during training. Beats Opus 4.7 on Terminal-Bench.
July 1, 2026 · 5 min read
⚡ SubQ: First Sub-Quadratic LLM with 12M Token Context
ArchitectureA startup claims to have escaped the quadratic attention bottleneck. 12M tokens, 64x less compute, frontier retrieval.
July 1, 2026 · 5 min read
🔬 Qwythos 9B: Claude Mythos Distillation
Fine-tuneA 9B model fine-tuned on Claude Mythos traces gains +34 MMLU. Apache 2.0, 1M context, native tool calling.
July 1, 2026 · 4 min read
🚀 DeepSeek DSpark: 60-85% Faster Inference
InferenceSpeculative decoding done right. No retraining, no quality loss. Already in production at DeepSeek.
July 1, 2026 · 5 min read
🐱 LongCat 2.0: Meituan's 1.6T MoE Stealth Model Revealed
New ReleaseThe mysterious 'owl-alpha' on OpenRouter was Meituan's LongCat-2.0 — 1.6T total params, 48B activated, trained on Huawei Ascend 910C.
July 1, 2026 · 4 min read