Gemini 3.8 Flash improves long-running coding and agent workflows while offering a one-million-token context window and adjustable thinking levels. This guide covers its specifications, benchmarks, pricing, tradeoffs, and migration requirements.
· 6 min read
Hermes Bot Mode is a desktop roster of named specialist agents. Each Bot is a Hermes profile with its own model, memory, skills, and chat. Bundled in v0.20.3. On by default. Not a new product.
· 3 min read
Ox Alpha is a stealth coding model, not a named lab launch. OpenCode made it free for a week with a 1M-token window; OpenRouter lists it as stealth/ox-alpha. The maker is still undisclosed.
· 3 min read
Computer History is an opt-in macOS ChatGPT desktop feature for Pro, Business, and Enterprise. It turns app and website interaction events into a timeline and local memories. Off by default. No screenshots. Memories required.
· 4 min read
When Claude Code forgets instructions or context fills fast, run /context before /compact. Check CLAUDE.md, auto memory, MCP, logs, and whether this is still one task.
· 7 min read
fx is Vercel Labs’ experimental Zig coding-agent CLI. About 6.39 MiB. Apache-2.0. Model-agnostic. Unix-shell form, not a terminal IDE. macOS and Linux.
· 3 min read
OCR 4.1 is Mistral’s July 16, 2026 public-preview OCR model. It keeps OCR 4’s bounding boxes and block labels, and adds block-level confidence scores. The API is $4 per 1,000 pages.
· 3 min read
gpt-image-2 can emit PNG or WebP with a real alpha channel. Preview in the API. background="transparent". The prompt still wins: if you describe a scene, you may get a scene.
· 2 min read
OpenSenseNova’s U1.5 is one native model for 4K generation and instruction editing. The Preview landed July 31. The current checkpoint is SenseNova-U1.5-8B-MoT, Apache 2.0, 18B on Hugging Face.
· 3 min read
msitarzewski/agency-agents is a MIT-licensed catalog of 230+ Markdown specialist personas. Install a subset into Claude Code, Cursor, Codex, and other tools, or use the files as a prompt library.
· 2 min read
Gemini Spark is Google’s 24/7 personal agent in the Gemini app. It runs tasks in the cloud after you close the laptop. Personal Google Account, 18+, AI Pro or Ultra, Keep Activity on. Not on work or school accounts.
· 5 min read
Meshy turns a text prompt or an image into a 3D mesh you can texture, remesh, rig, animate, print, or call from an API. Credits, not a one-click replacement for Blender.
· 3 min read
Odysseus is a self-hosted AI workspace: chat, agents, documents, email, notes, calendar, and Deep Research on your machine. AGPL. Docker on port 7000. Local models stay local; API models do not.
· 2 min read
Snapr is a macOS and Windows app for screenshots, scrolling capture, 4K 60fps recording, annotation, and light video edit. The current site puts every listed feature on the free plan. $9.99 lifetime is a supporter purchase.
· 4 min read
xAI’s August 12 flagship keeps Grok 4.5’s $2 / $6 base API rates, adds a 500K context window and xhigh reasoning, and posts mixed agent scores against GPT-5.6 Sol and Fable 5.
· 4 min read
Qwen’s August flagship is two products, not one. Qwen3.8-Max is the 2.4T Cloud API. The open-weight twin, Qwen3.8-2.4T-A95B, is text-only. A separate 27B vision model is the one you can actually run locally.
· 3 min read
DeepSeek’s flagship left preview on August 13. The GA build is deepseek-v4-pro (V4-Pro-0813): a 1-million-token, open-weight model with a higher API price and a published agent-score jump over the April preview.
· 3 min read
A Perseid peak is the densest part of a debris stream, not a switch that turns the shower off. In 2026 the maximum over Korea fell in daylight, but the stream is still open through August 24.
· 4 min read