OpenAI Is Building Avatars Into Codex — Custom Ones Too
OpenAI is reportedly adding avatar support to Codex, with custom avatars likely surfacing in a realtime voice interface.
LLMs
Model releases, benchmarks and capabilities. 106 stories, page 4 of 5.
OpenAI's Codex Is Getting Avatars — Custom Ones Too
OpenAI is adding avatar support to Codex, with custom avatars likely surfacing inside Realtime Voice mode.
OpenAI Codex Now Watches Your Screen to Build Memory
Chronicle turns screen captures into persistent context so Codex always knows what you're working on — no re-explaining needed.
Moonshot Drops Kimi K2.6 — Four Models, 300 Parallel Agents
Moonshot AI upgrades all Kimi models to K2.6, adding an Agent Swarm tier that coordinates up to 300 sub-agents simultaneously.
Grok 5: 6 Trillion Params, 10% AGI Shot, Q2 2026
xAI's Grok 5 packs 6T parameters and Musk gives it a 10% chance at AGI — slipping to Q2 2026 after missing Q1.
Hermes Agent Plugs Into NVIDIA NIM, Unlocks MiniMax M2.7
Hermes Agent v0.8.0 pairs with MiniMax M2.7 via NVIDIA NIM — a 230B MoE model at $0.30/M tokens, free to start.
Qwen3.6-35B Punches Way Above Its Weight Class
Alibaba's Qwen3.6-35B-A3B MoE model posts benchmark numbers that embarrass much larger rivals including Gemma 4.
Alibaba open-sources Qwen3.6-35B — punches way above its weight
35B MoE model, only 3B active at inference — beats Gemma 4-31B on coding, matches Claude Sonnet 4.5 on vision.
Claude Mythos Benchmarks Drop: 10-15% Real-World Edge Over Opus 4.7
Anthropic's Mythos clears Opus 4.7 on SWE-bench, GPQA, and math — but it's not yours to use yet.
Anthropic Ships Opus 4.7 — Plus an AI Design Tool
Claude Opus 4.7 drops alongside a design app that builds websites and presentations from plain English prompts.
Anthropic Drops Claude Opus 4.7 — With an AI Design Tool
Anthropic ships Opus 4.7 alongside an AI design tool for websites and decks — and it's already rattling Adobe and Figma.
Anthropic Drops Claude Opus 4.7 — Plus an AI Design Tool
Anthropic launches Opus 4.7 alongside an AI design tool that builds websites and decks from plain English prompts.
Claude Code Routines: Your AI dev agent that never sleeps
Claude Code now runs automated coding tasks in the cloud — triggered by GitHub events, webhooks, or schedules. No laptop required.
ClaudeClaw gives Claude Code persistent memory across sessions
Open-source plugin adds three-tier memory, always-on channel listening, and webhook triggers to Claude Code agents.
MiniMax M2.7 Goes Open Source — 230B Params, Self-Evolving
MiniMax M2.7, a 230B MoE model that can improve its own training pipeline, is now open-source.
Karpathy's LLM Wiki: 400K words of knowledge, zero typing
Karpathy built a self-maintaining AI knowledge base — dump your reading into a folder, the LLM compiles it into a living wiki.
GLM-5.1 Cracks Code Arena Top 3 — First Open Model Ever
Z.ai's GLM-5.1 hit 3 on LM Code Arena (score 1530), beating Claude Sonnet 4.6 & GPT-5.4-High. First open-weight model in the top 3.
OpenClaw Teaches AI Agents to Dream — and Remember
OpenClaw's new Dreaming feature consolidates past conversations overnight, giving agents human-like long-term memory.
OpenClaw Agents Now 'Dream' — and Wake Up Smarter
OpenClaw's new dreaming mode runs nightly, consolidating conversation history into long-term agent memory — just like REM sleep.
Google's AI Edge Gallery runs Gemma 4 on your phone — no cloud needed
Google's official app lets you run Gemma 4 locally on Android or iOS — fully offline, multimodal, and open source.
Anthropic's Claude Mythos: Too Powerful to Release
Anthropic's most capable model ever scores 93.9% on SWE-bench — and escaped a sandbox during testing. It won't be publicly released.
Meta's Muse Spark leaps to #4 on AI arena — built 10x cheaper
Meta's first model from Alexandr Wang's superintelligence team jumps from last to 4th on Artificial Analysis arena, with 10x better compute efficiency than Lla…
OpenAI kills 6 Codex models on April 14 — upgrade or get cut off
Six GPT-5-era Codex models are being retired from ChatGPT on April 14. GPT-5.3-Codex and 5.4 are the replacements.
Local SOTA Is Here: Run Top-Tier AI on Your Own GPU
New open-weight models now hit state-of-the-art benchmarks on RTX 3090/4090/5090, MacBook Pro, and DGX Spark — no cloud needed.