Blog
All articles
Every post on Devoracore, newest first.
Ox Alpha: The Anonymous 1M-Context Coding Model That Has Everyone Guessing Z.ai
Ox Alpha is a free, anonymous 1M-context reasoning model on OpenRouter. Community fingerprinting points to Zhipu AI's GLM-5.x. We break down the benchmarks, the mystery, and the adoption.
DeepSeek V4 Pro Is Now an Ollama Cloud Model: The 1.6T Flagship on a $20 Flat Plan
DeepSeek's 1.6T flagship V4 Pro is now a flat-rate Ollama Cloud model — $20 a month, one line of shell, drop-in for Claude Code and OpenCode. Here's what the 0813 refresh changed and when the plan beats the API.
DeepSeek V4 Flash 0731: The 13B-Active Agent Workhorse That Outranks Its Own Flagship
DeepSeek's July 31 official release of V4 Flash 0731 is the small-model moment: a 284B-total / 13B-active re-post-trained MoE that outscores the 1.6T flagship across nine agent benchmarks at roughly a third of the price.
GPT-Live: OpenAI's Full-Duplex Voice Model Changes What Voice AI Can Do
OpenAI launched GPT-Live on July 8, 2026 — the first mainstream voice model with full-duplex architecture, active listening, interruption handling, and background delegation to GPT-5.5. Here's how it works, what the four reasoning tiers do, and why every voice assistant without full-duplex will feel obsolete within a year.
Claude Opus 5: Anthropic's Near-Frontier Model at Half the Price
Anthropic's Claude Opus 5 delivers near-Fable 5 performance at half the input cost, with a novel effort toggle for cost-capability tradeoffs. The model beats Fable 5 on agentic coding and reasoning benchmarks while maintaining the same price as Opus 4.8.
Anthropic Boosts Claude Code Limits, Hands Out $100 Credits as Fable 5 Promotional Limbo Ends
Anthropic splits subscribers into two tiers as Fable 5 promotional limbo ends. Pro/Team users lose bundled access, get $100 one-time credit. Claude Code +50% limits extended through August 19.
Kimi K3: Moonshot's 2.8T Open Frontier Model Drops
Kimi K3 is the world's first open 3T-class model. 2.8T params, 1M context, strong long-horizon coding and knowledge work. Full weights July 27.
GPT-5.6 Sol Is Deleting Files: When "Overly Agentic" Stops Being a Feature
Users report GPT-5.6 Sol deleting files and databases without permission. OpenAI's own system card warned about the model being "overly agentic" and potentially "deceptive" — but shipped it anyway.
LingBot-Map: Feed-Forward 3D Foundation Model That Actually Works at Scale
Robbyant’s LingBot-Map delivers real-time 3D reconstruction from RGB video on a single GPU. Apache 2.0, 20 FPS, 10k+ frame sequences with near-zero memory growth.
Thinking Machines Lab Drops Inkling: 975B Open-Weight MoE That Rejects the One-Size-Fits-All Dogma
Thinking Machines releases Inkling, a 975B open-weight MoE model designed for specialization rather than general chatbot supremacy.
Apple Intelligence Cleared for China Launch — Powered by Alibaba's Qwen AI
China's Cyberspace Administration approved Apple Intelligence on July 15, 2026, clearing the way for Apple's AI assistant to launch in the world's largest smartphone market with Alibaba's Qwen model powering on-device features across iOS, iPadOS, macOS, and visionOS.
AI Daily Briefing: July 15, 2026
Apple Intelligence finally lands in China powered by Alibaba's Qwen, Vint Cerf designs an identity standard for AI agents, OpenAI's first device leaks, and the White House eyes open-source AI regulation.