来自 Swfte 团队
关于 AI、企业自动化和工作未来的最新思考。
AI is shifting from raw IQ to cost per unit of intelligence. Why efficiency, not size, is the race that matters now.
Fable 5 leads at $10/$50; GPT-5.6 sits two points back at half the price. Why the gap is really an efficiency story.
How models are learning to think in a compressed private notation instead of English — and why that quietly lowers cost.
Reasoning models bill you for hidden thinking tokens you never see. Here's the real cost, and how GPT-5.6 cut it.
GPT-5.6 matches Opus 4.8 on the AA Index at a third of the price. What changed, and whether it's worth switching.
DeepSeek V4.5 ships under MIT at $0.50/$1.10 with frontier-band reasoning. The numbers, the caveats, and the fit.
Gemini 3.2 Pro keeps 2M context and $2/$12 pricing while closing the quality gap. Where it wins and where it trails.
Llama 5 is the first Llama to rival the proprietary top tier on reasoning. What changed, and who it's for.
Claude Sonnet 5 brings Opus-class coding to $3/$15. Where it holds up, where it doesn't, and who it's for.
A benchmark testing agents on real pro workflows across 55 sub-industries, plus an open memory harness. Why both matter.
Claude Fable 5 launched and vanished in six days. The real lesson: single-vendor fragility and behavior observability.
Nine releases push generative AI off the flat image into 3D, 4D, motion transfer, and long-form video.
This week's four mostly-open primitives: diffusion text, voice-preserving translation, tiny TTS, open image training.
Four open-weight labs shipped frontier-adjacent models in a week: Kimi K2.7, MiniMax M3, GLM-5.2, Nexus N2.
Three releases teaching AI how the physical world moves — robot bodies, deformable objects, first-person human motion.
Kimi K2.7 Code: an open, Modified-MIT coding model good enough to self-host. Why open weights are winning in-house dev.
Anthropic Claude Fable 5 (June 9 2026): SWE-bench Pro 80.3%, $10/$50, now #1 on our AI model leaderboard.
MiniMax M3 (June 1 2026): open-weight, 1M-context, multimodal. SWE-bench Pro 59.0%; ranked on our leaderboard.
Claude Opus 4.8 retook #1 on the AA Intelligence Index at 61.4. What changed, and the 3 jobs GPT-5.5 still wins.
每周获取关于企业 AI 的见解，直接发送到您的收件箱。
无垃圾邮件。随时取消订阅。