LearnSphere

Kimi K3 is seriously impressive when it comes to 3D tasks. The quality, consistency, and overall performance make it one of the most exciting models to watch in this space.

• Better performance, reliability, and overall polish More improvements are on the way.

The latest ChatGPT desktop app update is here, shaped by user feedback. • Conversation history & Projects are back in the sidebar • Chat & Work history now sync across web, mobile, and desktop • Easily switch between Chat and Work modes • Codex mode remains unchanged

With all the excitement around Kimi K3, it's now available for OpenCode Go users. For now, it consumes usage limits at a higher rate since we haven't secured discounted pricing yet—but we're working on it and hope to improve this soon.

AI should empower people with more freedom, opportunity, and prosperity—not fear. That’s the future worth building.

The past 12 months weren’t our strongest, and I take responsibility for that. But what’s ahead is shaping up to be our best yet. The team is building incredible things, and I can’t wait for everyone to see what’s coming.

🚨 China has open-sourced a real-time AI model that reconstructs 3D scenes from a single video—no LiDAR needed. → ~20 FPS on one GPU → Stable across 10,000+ frames → Works with drones, driving, and indoor videos 100% open source. 🚀

What are the best harnesses to use with Kimi K3 right now? I've heard it's especially strong at orchestration. Which tools or frameworks showcase that best?

I usually ignore rumors, but if Kimi K3 really outperforms Opus 4.8, that's a huge surprise. Also hearing whispers that Opus 5 could be on the way. 👀

I talk to ChatGPT more than I type these days. The new voice model has crossed a real threshold—it finally feels natural enough to become my default way of interacting.

🚨 BREAKING: China's Kimi K3 has claimed the #1 spot on the Frontend Code Arena, overtaking Claude Fable 5 in the latest rankings. A major milestone in the AI coding race. 👀

We've removed the 5-hour limit for Codex Plus and Pro over the past few days. How has your experience been? Is the weekly usage limit easier to manage, or do you still find it restrictive? If you could redesign the system, what would your ideal usage model look like?

Build an artifact once, then it pulls live data via MCP. Artifacts use the viewers’ own MCP connectors, so creators don’t need to re-run sessions to refresh, and viewers can only see the data they have access to.

Claude Code artifacts can now use MCP connectors, enabling dashboards and apps to fetch data and perform actions for each viewer on demand. Available on Pro, Max, Team, and Enterprise plans.

GPT-5.6 Sol is half the price and often 2× more token-efficient than Fable for the same tasks—delivering comparable results at roughly one-quarter of the overall cost.

📱 GPT-Live now supports Live Activities on iOS. Enable Background conversations in Settings → Voice to use it. This will likely become the default soon.

You could spend $235 on OpenAI's programmable keyboard with a knob... Or grab a programmable keyboard with a knob on Amazon for around $18.

Cut GPU utilization by ~85% with a surprisingly simple fix. Neither Fable nor GPT-5.6 found the root cause—they focused on the wrong parts of the code. But GPT-5.6 did help by generating a debugging script that sped up the investigation.

A ChatGPT automation with read-only email access uncovered nearly $45K in billing errors across 3 years of construction invoices. Duplicate charges, incorrect amounts, and misaddressed invoices—all verified and resolved. A great example of AI delivering real ROI.

Thought I was about to announce a reset? Not this time. Just scrolling through X, looking for feedback on ChatGPT Work. What should we improve?