GranularGranular Hosted models
Use Claude, GPT, Gemini and more in Granular with no AI subscription or API key: the AI Add-on gives you hosted models paid for with credits.
Updated 2026-10-20
The AI Add-on (Granular Hosted) gives you Claude, GPT, Gemini and many more models inside Granular with no AI subscription to manage and no API keys to paste. You pay Granular, Granular pays the model companies, and a credit balance tracks what you have used.
What Granular Hosted is
Granular needs an AI model behind it to do any work. Usually you supply that yourself: you sign in with your own Claude or Codex account, or paste your own API key. That works well, but it assumes you already have one.
Granular Hosted removes that assumption. It is a paid add-on where Granular runs the AI connection for you:
- No separate AI subscription. You don't need a Claude or ChatGPT plan.
- No API keys. Nothing to create, paste or keep safe.
- One bill. Model use appears on your Granular bill, not on several AI companies' bills.
- Many models, one switch. Change model mid-conversation without signing in to anything.
It works on any plan, including Free. It is separate from the Pro plan: Pro lifts the limits on projects, sessions and agents, while the AI Add-on supplies the models.
New to Granular with no AI subscription? You can start with free AI credits during setup, no card needed.
The models, and what each is good for
The hosted lineup is 32 models from 9 companies: the leading AI labs plus strong challengers. Every one of them is agent-capable, meaning it can reliably use tools, which is what Granular needs. Models that are only good at chatting are left out on purpose.
Below is the same information the app shows when you hover over a model in the picker: strengths scored out of 5 for coding, reasoning, writing and everyday tasks, and the cost of a typical turn in credits. (A turn is one request and the agent's work on it.) The typical turn used here is a deliberately cautious estimate. Real turns often cost less.
Anthropic
| Model | Good for | Coding | Reasoning | Writing | Everyday | Credits per turn |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | The most capable model: elite coding and reasoning, and cheaper than Opus 5 | 5 | 5 | 5 | 2 | about 10 |
| Claude Opus 5 | Elite coding and reasoning, at flagship price | 5 | 5 | 5 | 2 | about 13 |
| Claude Fable 5.1 | The newest Fable: flagship-class, with the strongest long-form writing | 5 | 5 | 5 | 2 | about 13 |
| Claude Fable 5 | As capable as Opus 5, with an extra edge for long-form writing | 5 | 5 | 5 | 2 | about 13 |
| Claude Opus 4.8 | Last-generation flagship, still top-tier for coding and writing | 5 | 4 | 5 | 2 | about 13 |
| Claude Opus 4.7 | Earlier Opus 4 flagship | 4 | 4 | 4 | 2 | about 13 |
| Claude Opus 4.6 | Earlier Opus 4 flagship | 4 | 4 | 4 | 2 | about 13 |
| Claude Sonnet 4.6 (default) | The balanced default: much of Opus 5's quality, far cheaper | 4 | 4 | 4 | 4 | about 8 |
| Claude Sonnet 5.5 | The newest Sonnet: fast, balanced, at half Opus 5.5's price | 4 | 4 | 4 | 4 | about 5 |
| Claude Sonnet 5 | Previous-generation Sonnet: a strong daily driver at a fraction of Opus 5's cost | 4 | 4 | 4 | 4 | about 5 |
| Claude Haiku 4.5 | Fast everyday model: quick chores over hard reasoning | 3 | 2 | 3 | 5 | about 3 |
OpenAI
| Model | Good for | Coding | Reasoning | Writing | Everyday | Credits per turn |
|---|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI's most capable, built for long, complex work | 5 | 5 | 5 | 1 | about 25 |
| GPT-5.6 Sol | Frontier coding: strongest on multi-step agent work | 5 | 4 | 3 | 3 | about 5 |
| GPT-5.6 Terra | Balanced everyday model at a mid-tier price | 4 | 4 | 4 | 4 | about 6 |
| GPT-5.6 Luna | Fast and affordable: high-volume chores and light agent work | 3 | 3 | 3 | 5 | about 1 |
| GPT-5.5 | Frontier general model: strong reasoning on hard professional work | 4 | 5 | 4 | 2 | about 14 |
| GPT-5.1 | OpenAI's previous flagship | 4 | 5 | 4 | 3 | about 4 |
| GPT-5.1 Codex | OpenAI's previous best coder | 5 | 4 | 3 | 2 | about 4 |
| GPT-5.1 Codex Mini | A cheap, capable coder | 4 | 3 | 2 | 4 | about 1 |
| GPT-5 Mini | Cheap, fast general model for everyday chores | 3 | 3 | 3 | 5 | about 1 |
| Model | Good for | Coding | Reasoning | Writing | Everyday | Credits per turn |
|---|---|---|---|---|---|---|
| Gemini Pro | Google's flagship: top-tier reasoning and huge memory for long documents | 4 | 5 | 4 | 3 | about 6 |
| Gemini 3.8 Flash | Google's newest Flash: better coding and agent work, still cheap | 4 | 4 | 3 | 5 | about 2 |
| Gemini 3.5 Flash | Near-Pro coding at Flash speed | 4 | 3 | 3 | 3 | about 4 |
| Gemini 3.5 Flash-Lite | Lightest current Gemini: focused small tasks | 3 | 2 | 3 | 5 | about 1 |
| Gemini Flash | Fast, cheap and capable: a great everyday workhorse | 3 | 3 | 3 | 5 | about 1 |
| Gemini Flash-Lite | The lightest, fastest option: quick chores | 2 | 2 | 3 | 5 | about 1 |
DeepSeek, Kimi, GLM, Grok, Qwen and Mistral
| Model | Good for | Coding | Reasoning | Writing | Everyday | Credits per turn |
|---|---|---|---|---|---|---|
| Grok 4.6 (xAI) | A well-rounded, up-to-date generalist | 4 | 4 | 4 | 3 | about 4 |
| DeepSeek v3.1 | Strong open coder and reasoner, at a fraction of the cost | 4 | 4 | 3 | 4 | about 1 |
| Kimi K2 Thinking (Moonshot) | Capable reasoning over long documents at a very low price | 4 | 4 | 3 | 3 | about 1 |
| GLM-4.6 (Z.ai) | Strong coding agent at a fraction of the cost | 4 | 3 | 3 | 4 | about 1 |
| Qwen3 Max (Alibaba) | Strong multilingual coding, low cost | 4 | 4 | 3 | 3 | about 2 |
| Mistral Medium 3.1 | A solid all-rounder for writing and everyday work | 3 | 3 | 4 | 4 | about 1 |
How the model list is laid out
In the model button under the message box, the AI Add-on's models are grouped by maker: Anthropic, OpenAI and Google first, then the other makers. Each group shows its newest models first. Older versions of a model are folded under Previous generation · N; click it to open. In the Anthropic group, that's where you'll find Claude Opus 5, Opus 4.8, 4.7 and 4.6, Fable 5, Sonnet 5 and Sonnet 4.6.
Type in the search box to find a model by name or maker, and pin the ones you use most so they stay at the top (pin icon, or Cmd+D / Ctrl+D). Each row shows roughly how many credits a typical turn uses. See Pick and pin models.
Not sure where to start? Use Claude Sonnet 4.6. It is the default because it balances quality and cost well. (Because Sonnet 5.5 is newer, you'll find Sonnet 4.6 under Anthropic's Previous generation.) Reach for a flagship when a task is genuinely hard, and drop to a 1-credit model for chores. You pick the model for each message, so switching costs nothing but a click, and hovering any model shows its card before you choose.
How credits work
Hosted use is measured in credits, and the rule is simple:
100 credits = $1 of model use, at the AI company's own price.
A credit buys the same amount of real work whichever model you spend it on. A more expensive model simply uses credits faster, in proportion to what it really costs.
Your balance has two parts, and they behave differently:
- Monthly credits come with your tier and reset at the start of each billing period. Unused monthly credits do not carry over.
- Credit pack credits are bought separately and roll over until you use them.
Granular always spends your monthly credits first, so the credits that don't expire are the last to be touched.
Tiers and credit packs
The AI Add-on is billed monthly. The only difference between the tiers is how many credits arrive each month. Every tier unlocks the same models.
| Tier | Price | Credits each month |
|---|---|---|
| Base | $20/month | 1,600 |
| Plus | $50/month | 4,000 |
| Max | $100/month | 8,000 |
Need more in a given month? Credit packs are one-time purchases whose credits roll over until spent: $10 for 800 or $25 for 2,000.
Two things to know when choosing. First, every tier and pack buys credits at the same rate. There is no bulk discount, so choose based on how much you'll really use. Second, remember that monthly credits reset each billing period while pack credits keep. If your use is uneven, a smaller tier topped up with packs in busy months often fits better than a big tier you only sometimes fill.
You'll find both in Settings → Billing & plan. If you're signed in when you buy, the purchase is added to your account automatically. You'll also get an unlock code by email as a backup. To move between Base, Plus and Max later, use Change plan on the AI Add-on in the same place. See Plans and pricing.
Watching your balance
The credits gauge sits beside the model picker, under the message box, and updates as you work. For the fuller picture:
- The gauge shows remaining credits at a glance, without leaving the conversation.
- Settings → Billing & plan shows your tier (marked Your tier), when it renews, and the email it bills to.
Counting credits happens in the background after your answer has arrived. It never slows down or interrupts a conversation.
When credits run out
Nothing breaks and nothing is lost. Hosted messages simply pause until there is balance again. Your sessions, files, terminals and history are untouched.
You have three ways forward:
- Top up or move up a tier if hosted models are how you want to work.
- Switch to your own Claude or Codex sign-in if you have one. Work continues immediately.
- Switch to your own API key and pay the AI company directly. See Connect an AI model.
Granular Hosted is a convenience, not a lock-in. The app works fully on an AI connection you own.
Using your own model instead
Switching away from Hosted takes one change in the model picker:
- Connect an AI model: sign in with Claude or Codex, or add an API key.
- Connect Codex: use OpenAI's Codex.
- Connect Gemini: Google Gemini with a free API key.
- Add a custom AI provider: any compatible AI service.
You can keep the AI Add-on and still use your own sign-in day to day. The picker works per conversation, so nothing forces you into one or the other.