ChatGPT Images 2.0
Article TL;DR: OpenAI’s new image model nails text rendering and photorealism, at 2x Gemini’s price. Key Takeaways High-quality 1024x1024 images cost $0.211 vs $0.167 for Images 1—prices rose Text ren…
Top stories from Hacker News — for AI builders, hackers, and founders.
Article TL;DR: OpenAI’s new image model nails text rendering and photorealism, at 2x Gemini’s price. Key Takeaways High-quality 1024x1024 images cost $0.211 vs $0.167 for Images 1—prices rose Text ren…
Article TL;DR: SpaceX paid $10B for services plus an option to acquire Cursor at $60B—not a straight buyout. Key Takeaways Bloomberg: $10B services deal now, option to acquire at $60B later this year …
Article TL;DR: Meta will screenshot and log employee keystrokes on work devices to train computer-use AI models. Key Takeaways Captures input from Gmail, Google Docs, and social media with no personal…
Article TL;DR: GitHub Copilot yanked Opus 4.6 mid-cycle, raised Opus 4.7 to 7.5x credits—no notice, no refunds. Key Takeaways Opus 4.7 costs 7.5x credits in Pro+; effective Opus price roughly doubled …
Article TL;DR: Brex open-sourced an LLM-as-judge HTTP proxy to gatekeep what production AI agents can call. Key Takeaways Policies auto-generated from real traffic snapshots, then validated against he…
Article TL;DR: Zig-based agent browser claims 16% lower token cost than Vercel’s tool, tiny install footprint. Key Takeaways Self-published single benchmark on Google Flights task; not independently v…
Article TL;DR: Anthropic silently removed Claude Code from Pro docs—confirmed A/B test on 2% of new signups. Key Takeaways A/B test targets ~2% of new prosumer signups; existing Pro and Max subscriber…
Article TL;DR: Go-compiled LLM gateway alternative to Python-heavy LiteLLM, with supply chain security advantages. Key Takeaways Compiled Go binary reduces runtime supply chain attack surface vs Pytho…
Article TL;DR: Stateful diagram runtime for agents via structured operations protocol—proprietary SaaS, no open source. Key Takeaways Diagram Scene Protocol enables structured mutations vs one-shot ge…
Article TL;DR: Functional shell, filesystem, and GPIO control crammed into 2KB RAM on an Arduino UNO. Key Takeaways Implements shell, filesystem, and GPIO CLI within hard 2KB RAM constraint Real-time …
Article TL;DR: Agents mimic human failure modes—ignoring constraints, improvising, then narratively defending bad output. Key Takeaways Agents drift toward training data averages when constraints conf…
Article TL;DR: CLI tool serializes agent session context to resume work across Claude Code and Codex harnesses. Key Takeaways Prompt cache doesn’t work cross-provider—every harness switch is a cold co…
Article TL;DR: Anthropic staff said OpenClaw CLI reuse is allowed again—verbally, no official policy published. Key Takeaways Anthropic still blocks parts of OpenClaw’s system prompt in practice despi…
Article TL;DR: Amazon invests $5B in Anthropic; Anthropic commits $100B back to AWS—effectively a 5% rebate deal. Key Takeaways Deal centered on Amazon Trainium chips; Anthropic gets $5B, commits $100…
Article TL;DR: Alibaba’s Qwen3.6-Max-Preview claims near-SOTA coding benchmarks, cloud-only, pricier than Kimi. Key Takeaways SWE-Bench Pro 57.3 vs Kimi K2.6’s 58.6; Qwen at $7.8/M out vs Kimi’s $4/M …
Article TL;DR: Coordinated fake star farms are systematically gaming GitHub to influence VC funding decisions. Key Takeaways VCs explicitly use GitHub stars as sourcing signals; median star count at s…
Article TL;DR: Claude Design runs on Opus 4.7 while Figma Make uses Sonnet 4.5—a permanent capability ceiling for Figma. Key Takeaways Figma pays Anthropic for Sonnet inference; Anthropic’s own produc…
Article TL;DR: Moonshot released a test suite to catch providers silently running cheaper quantizations of Kimi models. Key Takeaways AWS Bedrock silently drops 20-30% of Kimi K2 tool call attempts—ef…