The "which should I use?" chooser
For each task, the current best pick — with a one-line why and the date we last affirmed it. AI models and smart-home devices, side by side.
Newest isn't always best. A pick is the current choice as of its date — not whatever shipped most recently. Every pick is dated and tracked over time.
The current pick per task across the frontier and open-weight model families.
Gemini 2.5 Flash-Lite — $0.10/$0.40 per 1M with a 1M-token context; the budget workhorse.
GPT-5.4 nano — cheapest model with GPT-5-generation capability ($0.10/$0.40), for classification/routing.
Ministral 3B — the cheapest API option anywhere at $0.04/$0.04 per 1M; open weights.
Claude Opus 4.8 — reliability-focused flagship, 4x less likely than 4.7 to let its own code flaws pass.
Claude Sonnet 4.6 — Opus-level coding quality at $3/$15, 79.6% SWE-bench Verified.
Gemini 3.5 Flash — beats Gemini 3.1 Pro on coding benchmarks at ~25% lower cost.
The AI Image & Video section tracks the current image-generation leaders across providers.
xAI Aurora — image generation built into the Grok interface.
Qwen3-32B — leading open-weight (Apache 2.0) mid-size general model with a configurable thinking mode; runs locally.
Llama 3.3 70B — matches Llama 3.1 405B quality at a fraction of the cost; the community default for self-hosted text.
Mistral Small 4 — unified open-weight (Apache 2.0) reasoning + vision + coding model in one small package.
Llama 4 Scout — 10M-token context, the longest of any released model; best needle-in-a-haystack retrieval.
Gemini 3.1 Pro — 2M-token context, the largest production context window of any hosted API.
GPT-5.5 Pro — maximum-reasoning-depth variant for research-grade problems.
Gemini 2.5 Pro — proven Deep Think reasoning with a 1M-token context.
Grok 4.20 — extended-reasoning Think Mode with the strongest hallucination resistance in the Grok family.
The AI Image & Video section tracks the current video-generation leaders across providers.
GPT-4o — native audio/voice model; still the API workhorse for real-time voice tasks.
GPT-4o mini — cheap native-audio model ($0.15/$0.60) for budget voice apps.
Claude Opus 4.8 — the frontier Anthropic flagship most consistently preferred for long-form prose and editorial voice.
GPT-5.5 — strong all-round writer with a 1M-token context, close behind on draft quality.
The current pick per task across hubs, cameras, and the connected-home stack.
Voice-first hubs and presence automation make the smart home a powerful accessibility tool — start with the Smart Home guide.
Aqara Hub M3 — Matter controller + Thread border router + Zigbee 3.0 in one box; the lowest-friction multi-ecosystem hub.
Amazon Echo Hub — purpose-built dashboard with Matter + Zigbee + Thread baked in for Alexa households.
SmartThings Station — combo Qi charger + Matter / Thread / Zigbee hub at a small footprint.
Picks are editorial judgements, each stamped with the date it was last affirmed. Disagree? Tell us — the point is to keep these current.
Nest Cam (indoor, wired) — clean privacy story and a legible app at an affordable price.
Aqara Camera Hub G3 — 2K pan/tilt camera that doubles as a Zigbee hub, with local face recognition and HomeKit Secure Video.
