Frontier reasoning models
The slow, careful brains. Used when an answer has to defend itself — long arguments, multi-hop research, code that has to be right the first time.
Models
The right model
for the job. Always.
askFinz quietly picks the best AI for what you're doing. Different work, different brains. You don't need to know which one — but if you want to know, we tell you. Every answer carries a small badge naming the model behind it. Every choice is one click to change.
Why we choose for you
Calibration over capability.
The most capable model on the leaderboard is rarely the best one for what you're actually doing. It depends on the shape of the question — its length, its certainty, the cost of being wrong. Picking is a calibration problem, not a ranking problem. We'd rather pick well than pick big.
You won't see us crown a single AI as “the askFinz brain.” That's the move that makes products stale.
Providers
07 familiesWe work across families instead of crowning one. Each family earns its turn on the kind of work it's actually best at — and we tell you which one wrote each answer.
Frontier chat models
Conversational, fast, broadly capable. The default for everyday questions and drafting where pace matters as much as polish.
Open-weight providers
Open-weight families you can self-host or pin. Useful when cost, residency or independence from any single vendor is a hard constraint.
Vision-capable models
Read a screenshot, a chart, a photographed whiteboard. Words and pictures handled in the same turn — useful for review, extraction and accessibility.
Code-tuned models
Trained heavier on code than prose. Stronger at refactors, test generation, language-specific idioms and reading large unfamiliar codebases.
Long-context models
For when the question carries a hundred pages with it — the entire annual report, the merged transcript, the full prior thread.
Audio / TTS models
Read out a briefing, transcribe a meeting, draft a script that's meant to be heard. Voice in, voice out — without leaving the workshop.
Providers we route across
10 namesWe route through frontier providers, open-weight model families and specialised audio, vision and code stacks. Pick by quality, by cost, or let the router pick for you.
OpenAI
Gemini
Mistral
Llama
Qwen
DeepSeek
Moonshot
NVIDIA NIM
Z.ai
Google Cloud
Shown as monochrome approximations, not real brand assets. The live, per-seat model matrix lives at dash.askfinz.ai/usage/models (admin-only).
Hundreds more in the extension
380+ more modelsThe cloud models above are only half the story. With the askFinz browser extension you also get 380+ more AI models, ready to use right alongside the cloud ones.
askFinz surfaces the ones that fit, so there's always more choice close at hand. Pick one yourself, or let askFinz choose for you — the way it does with every other model.
See the extensionTwo ways to use AI
In the cloud · 10 leading models
The best-known AI, picked for each task — always with the choice to switch.
In the extension · 380+ more models
Hundreds more AI models, ready whenever you are.
Cloud models and extension models, side by side.
Model types
07 shapes of brain- ChatConversational answers, drafting, day-to-day questions.
- ReasoningSlow, deliberate, multi-step — the brains for hard answers.
- VisionReads images alongside words on the same turn.
- CodeEngineering work — refactors, tests, language-specific idioms.
- EmbeddingVectorises text for semantic search across your library.
- AudioSpeech-to-text, text-to-speech, briefings you can listen to.
- Image-genGenerates illustrations, charts and figures from a prompt.
The full live model catalogue — every model currently routable, with throughput and cost — lives at dash.askfinz.ai/usage/models for admins on your account.
Rules of thumb
05 of them- 01Long-form, careful work.
When the answer needs to hold up — an investigation, a memo, a literature review — askFinz picks a model that prefers care to speed. The output is slower, the citations are real.
- 02Code review and refactor.
Different work, different brain. We pick a model that reads more than it writes, then writes less than you'd expect — short, deliberate, with the rationale.
- 03Fast factual lookup.
When the question is small, askFinz picks a fast, cheap, quietly capable model. The answer arrives in a heartbeat. The honesty doesn't move.
- 04Reading what's on the page.
Vision-shaped work — a chart, a screenshot, a whiteboard photo — goes to a model built for it. Words and pictures, on the same turn.
- 05When you'd rather decide.
Pin a model in a workspace and we'll keep using it. Override per turn and we'll respect it. The router is sensible, not insistent.
Many minds, one mind.
We work with the leading AI labs and the open-weight ecosystem alongside them. You don't pick. We pick. The decision is always visible.
Visible by default.
Every answer carries a small badge naming the model that wrote it. Routing is intent-aware but never invisible — see /trust for the commitment.
Override, always.
Pin a model per workspace, override per turn, or hand us a key for a model you already pay for. We treat preferences as durable; we explain ourselves when we disagree.
How we choose, in plain English. The longer write-up lands on /research when the first notes go out.
How we keep ourselves honest