Blog
Notes from the team.
Model comparisons, product updates, and the occasional opinion on where multi-provider AI is going.
Subscribe via RSS.
- features canvas
Canvas: a live preview pane for everything a model writes
When a model writes HTML, SVG, a Mermaid diagram, or Markdown, you shouldn't have to copy it somewhere else to see it. Canvas renders it next to the chat and updates as you iterate.
Read Article →
- features search
Web search that cites its sources, on every model
Search in the studio isn't a separate mode you switch into. It's a tool any model can call mid-answer — and every fact it pulls comes back with the URL that informed it.
Read Article →
- features workflow
Branch the loser, keep the winner: model A/B inside one thread
Switching models mid-conversation isn't a gimmick. It's the fastest way to find out which frontier model is actually best at your specific task — without re-typing the prompt or losing the context.
Read Article →
- video release-notes models
HappyHorse 1.0: the first video model that nails sync audio
Alibaba's HappyHorse 1.0 shipped this week. The visuals are impressive. The thing nobody else has solved — joint audio-video generation — is what makes it the new default in our lineup.
Read Article →
- benchmarks comparisons models
GPT-5, Claude Opus 4, Gemini 2.5 Pro: a 2026 head-to-head
We ran the same 18 prompts through every frontier model and tracked where each one actually wins. The answer isn't 'use the newest one' — it's more interesting than that.
Read Article →
- video models release-notes
Seedance 2.0 and the multi-input moment for video
Nine reference images. Three reference clips. Three audio tracks. We've been trying to break Seedance 2.0 for a week — here's where it shines and where it still trips.
Read Article →
- pricing philosophy
Credits, not tokens: a pricing decision in the open
Why we ditched per-provider, per-token math and shipped a single shared credits pool. The honest argument, the math we ran, and the case against.
Read Article →
- philosophy reliability
Why multi-provider AI matters more than the model you use today
Single-provider lock-in feels fine — until the day your provider has an outage, ships a regression, or quietly rate-limits your account. The argument for keeping options open.
Read Article →
- video image release-notes
Grok Imagine in the studio: what it's actually for
xAI shipped a video + image model tuned for social. We added it to the lineup and worked out who should reach for it instead of HappyHorse, Veo, or FLUX.
Read Article →
- reasoning models benchmarks
Reasoning models in 2026: when extended thinking actually pays off
Opus thinking, GPT-5.4 Mini, DeepSeek R1 — three takes on the same idea. We tracked which problems get sharper with more compute, and which ones just get slower.
Read Article →
- image comparisons benchmarks
FLUX.2 Pro vs Nano Banana 2 for product shots: a side-by-side
Two top image models, one specific job. We ran 30 product-photography prompts through both and tracked which one we'd actually ship to a client.
Read Article →
- video release-notes
Adding video: how we picked the launch lineup
Eight video models, one shared credits pool. What made the cut, what didn't, and the four criteria we used to decide.
Read Article →
- onboarding tutorial
Getting started with Any AI Studio in five minutes
A walk through the shortcuts and small features that pay off the most in the first week — model switching, branching, Canvas, memory, and shared chats.
Read Article →