Cloud LLM Models
The catalog of hosted, cloud-based AI providers available under Settings → AI Models → Writer Engine → Cloud LLM (Hosted API).
Overview
Cloud LLM (Hosted API) sends your requests to a third-party AI provider's servers over the network — Anthropic, OpenAI, Google, DeepSeek, Cohere, Groq, or OpenRouter. Unlike Ollama or In-Browser Models, no model runs on your machine or in your browser tab; the provider's own infrastructure does the generation and streams the result back.
Providers are grouped here by architecture: Direct providers serve their own models from their own infrastructure; Proxy / Router and Serverless providers run open models (like Llama or Mixtral) on managed infrastructure; Aggregator providers (like OpenRouter) route to many providers behind one API.
Where to configure it
Go to Settings → AI Models → Writer Engine and select Cloud LLM (Hosted API). Pick a provider and model from the list below. See Agentic AI Settings for the full settings reference, and Ollama Integration, Local Language Models, or In-Browser Models for engines that don't require a cloud API key.
Screen preview
A quick look at the screens. Use the toggle in the corner of each preview to switch between the dark and light themes.
Claude Sonnet 4.6, Opus 4.7, Haiku 4.5
DeepSeek-V3, DeepSeek-R1
Llama 3.3 70B at record inference speed
GPT-4o, GPT-4o mini, o3
Gemini 2.5 Pro, Gemini 2.5 Flash
One key, hundreds of models
Command R+, Command A
Cloud LLM providers
Anthropic — Direct
Anthropic's own Claude family, called directly against Anthropic's API:
| Model | Type | Best for |
|---|---|---|
Claude Sonnet 4.6 | Direct | Balanced default — strong reasoning and writing quality at moderate cost. |
Claude Opus 4.7 | Direct | Anthropic's most capable model for the hardest reasoning and analysis tasks. |
Claude Haiku 4.5 | Direct | Fastest and cheapest Claude model for lightweight, high-volume tasks. |
OpenAI — Direct / Serverless
| Model | Type | Best for |
|---|---|---|
GPT-4o | Direct | General-purpose flagship model with strong multimodal support. |
GPT-4o mini | Direct | Cheaper, faster GPT-4o variant for everyday tasks. |
o3-mini | Serverless | Reasoning-focused model for step-by-step logic and math. |
Google AI Studio — Direct
Called directly against Google's own API — notable for aggressive pricing and very large context windows:
| Model | Type | Best for |
|---|---|---|
Gemini 2.5 Pro | Direct | Google's most capable model — large context window, strong reasoning. |
Gemini 2.0 Flash | Direct | Fast and inexpensive — calling it directly here is significantly cheaper than routing through an aggregator like OpenRouter. |
DeepSeek — Direct
DeepSeek's own models, called directly — widely regarded as the strongest cost/performance ratio in the current market:
| Model | Type | Best for |
|---|---|---|
DeepSeek-V3 | Direct | General-purpose chat and everyday tasks at very low cost. |
DeepSeek-R1 | Direct | Advanced reasoning and coding — includes an explicit "thinking" step. |
Cohere — Proxy / Router
| Model | Type | Best for |
|---|---|---|
Command R7B | Proxy / Router | Lightweight Cohere model for retrieval-augmented and enterprise tasks. |
Command R+ | Proxy / Router | Cohere's larger model for more complex retrieval and multi-step tasks. |
Groq — Serverless
Runs open models (Llama, Mixtral) on Groq's custom LPU hardware — the first choice for anyone who wants open-source models at record-breaking token-per-second speed:
| Model | Type | Best for |
|---|---|---|
Llama 3.3 70B (Groq) | Serverless | Meta's Llama 3.3 at Groq's very high inference speed. |
Mixtral 8x22B (Groq) | Serverless | Mistral's mixture-of-experts model, also at Groq speed. |
Modelion — Proxy / Router (Gateway)
Modelion is TotalApp's default gateway. Modelion Auto (recommended) lets the gateway pick the model; the other Modelion entries are read live from the gateway for your key, so the list always matches what your account can actually call.
Models that need your own provider key
Some models listed by Modelion (for example DeepSeek and GLM entries) are available only through your organization's own provider account. They are shown in the list but disabled, with the gateway's reason byok_binding_required next to them. Connect that provider's key in the Modelion console and the model becomes selectable.
Reasoning models can be slow
If Modelion Auto routes to a reasoning model, it thinks before answering, so a request can take a minute or more. TotalApp stops waiting for a gateway reply after about 30 seconds and reports Modelion request timed out. For quick structured answers prefer a non-reasoning model.
OpenRouter — Aggregator
Routes to many underlying providers behind one API — convenient for trying multiple models without separate API keys, at a small markup versus calling providers directly:
| Model | Type | Best for |
|---|---|---|
Llama 3.3 70B | Aggregator | Open-weight model via OpenRouter's routing. |
Gemini 2.0 Flash | Aggregator | Available here for convenience, but calling Google AI Studio directly (above) is cheaper. |
Mistral 7B | Serverless | Small, fast open model for lightweight tasks. |
Choosing a Provider by Architecture
Direct
Anthropic, OpenAI, Google AI Studio, DeepSeek — call the provider's own API directly. Lowest latency and cost for that provider's models; requires that provider's own API key.
Proxy / Router & Serverless
Cohere, Groq — managed infrastructure running either the provider's own models (Cohere) or open models on specialized hardware (Groq).
Aggregator
OpenRouter — one API key, many providers. Convenient for experimentation, but usually carries a small markup over calling a provider directly.
Tip: avoid paying the aggregator markup
If a model is available both directly (e.g. Google AI Studio's Gemini 2.0 Flash) and through OpenRouter, calling it directly is normally cheaper and faster — use the aggregator mainly for models that don't have their own direct listing here.
Selecting a Provider
Each provider requires its own API key configured on the server side — see Agentic AI Settings for where credentials are managed.