How engineers use AI, how to do it well, and what it means for your job. Written for people who ship and run the software, not people who build models.
PhoneBot from UCLA's Dennis Hong lab uses a $50-80 used smartphone as the robot's brain. Here is how it works, how it recovers from falls, and what it costs.
Perplexity Computer is reported to run on a Mac with Qwen 3.6 instead of cloud machines. What the 36GB unified memory figure means and what to check first.
Local LLM hardware tiers compared: RTX 4090, RTX 5090, RTX Pro 6000 Blackwell and Mac Studio, with prices, tokens per second, and multi-agent limits.
Flint is roughly 10x faster than dense Qwen 3.8 27B but scores 55 vs 73 on Terminal Bench. Here is where each local coding model fits.
Carbon-X is a 14.8B open DNA and text model that fits on one 80 GB GPU. Set up Transformers v5 or vLLM, handle DNA tokens, and call it from an agent.
Five habits for keeping Claude Opus 5.5 on track during hour-long coding runs: simple prompts, benchmarks, tests, reference designs, and team orchestration.
How to wrap an autonomous AI software factory in guardrails: agent and skill registries, a context lake, readiness scorecards, and controls, using Port.
Kuaishou's Kling AI video unit plans a Hong Kong IPO next year at about $18B. Kuaishou holds 68%, with Alibaba, Tencent and Baidu as minority backers.
InclusionAI's MIT-licensed Ling-3.0 Flash Santé is a 124B MoE healthcare model. See clinical test results, benchmark standing, and its lack of vision.
Flint runs about 10x faster than a dense 27B model. See how 2.5B active parameters, hybrid attention, a 4-bit output layer and a fused kernel do it.
Recursive, led by Richard Socher with $650M raised, says its Eureka machine helped design Flint's architecture, quantization, and kernels. What is known so far.
Wrap a vLLM-served Carbon-X as an agent tool: ask_carbon, tool calling, Claude Code and Codex skills, Inspect, and tips for passing DNA by reference.
Carbon-X, a 14.8B DNA model, averages 66.7% on five benchmarks versus 67.5% for GPT-6.1 Sol at far lower cost. Results, weak spots, and limits.
ByteDance's OmniHuman 2.0 turns one image and audio into up to five minutes of 720p video. Price fell from 12 to 7 cents per second, about $21 per full clip.
Fish Audio Drama 3 API costs about $10 per million characters at launch, then $50 after Nov 12. License limits, fallback billing and free trial explained.
Claude Duet is in research preview for Pro and Max subscribers in the US via waitlist. Here is who qualifies and how it shares your plan's usage limits.
Flint 40B scores 54.9% on Terminal-Bench 2.1 and 74.2% on SWE-bench Verified. See M4 Pro speed, NVFP4 vs BF16, and where Qwen 3.8 still wins.
Matrix Game 4.0 hands a video model a low-poly render. Nvidia StreamRender H3 hands it a color-coded map. Here is how the two approaches differ.
Sage-A1.5, a 35B MoE agentic model, hits 95.1 on GAIA and 61.2 on SWE-bench Pro. Here is where it beats or trails Kimi-K3, GLM-5.3 and GPT-5.6-sol.
Install Flint 40B locally with Ollama, llama.cpp or SGLang. Covers VRAM needs (about 32GB for the quant), GGUF Q4, NVFP4, MLX and BF16 on H200 or B300.
Prompt rules are not spending limits. Learn how a virtual card with a weekly cap, category and country locks, expiry and webhooks constrains an AI agent.
Sage-A1.5 can fan work out to parallel subagents through a spawn_subagent tool and self-review with Verification Mode. Setup, schema and harness notes.
Serve Sage-A1.5 with SGLang or vLLM: standard and MTP launch commands, parsers, and the YaRN factor 2.0 setup that extends 256K context to 512K.
Hands-on look at Vast-10M, which swaps attention in DeepSeek V4 and GLM 5.2 for VSA. Results from a game build, HTML art, a legal corpus and 81 languages.