maxwell ⩇⩇:⩇⩇

We've productionized query-aware compression for faster, cleaner, more-accurate search. Better context is better than more context. Our system cuts context tokens up to 70% while improving answer quality.

I think Gemini Omni just passed my first “does this fall apart when it moves?” test. The prompt was simple: a person working on a laptop, NYC at night in the background, reflections on the glass, hands moving, screen glow, and a camera shift. That’s a lot of little things that can go wrong. The result was honestly impressive.

My AI research workflow is turning into a small stack. Perplexity for source discovery. GPT-5.5 Instant for the fast scan. GPT-5.5 Thinking for the actual read. Gemini for the caveat check. Human review before posting. This run caught something useful: the draft missed Gemini Embedding 2 and included one line the source did not support. That’s why the handoff matters. The goal is not one perfect model. It’s a workflow that catches mistakes before they become posts.

gpt-5.5 instant becoming the default model is a big shift most users won't care about benchmarks they'll care about two things: - does chatgpt feel more natural? - does it remember useful context without getting weird? openai is clearly moving chatgpt beyond a normal chatbot