AI Models
Last updated: May 26, 2026
Credit rates
Novis uses a credit system for AI and transcription usage. Plans include a fixed monthly credit allowance based on your subscription tier. The same credit volume applies whether you bill monthly or annually — annual billing receives a ~17% discount on subscription price (2 months free).
Rates
USD is the reference currency; EUR and GBP values are shown for reference based on the applicable exchange rate.
| Currency | Per credit | Transcription |
|---|---|---|
| USD | $0.0025 / credit (400 credits = $1) | 500 credits / hour |
| EUR | €0.00225 / credit (400 credits = €0.90) | 500 credits / hour |
| GBP | £0.0020 / credit (400 credits = £0.80) | 500 credits / hour |
AI chat and Tables usage draws on credits according to the tables below (Primary Models).
Novis allow Authorised Users to choose and use third party AI models below (each, a "Model") to process customer data on behalf of Novis Customers and in accordance with contract terms between Novis and the Customer.
Customer can enable or disable any specific Model to make it available to its Authorised Users, evaluating at its own discretion and risk compliance with regulatory requirements (e.g. GDPR).
By default, the platform enables Models that align with the Customer's selected Data Residency Zone requirements. This includes Models hosted within the zone or those provided by Sub-processors ensuring adequate data protection safeguards (e.g., EU-US Data Privacy Framework, Standard Contractual Clauses). Customers may optionally enable other Models, acknowledging that data may be processed in regions with different privacy standards.
How Calculation Works
Usage is measured in Pages, where 1 Page = 640 Model Tokens (T) (approx. 500 words).
Costs are calculated in Credits (C) based on the model used.
The Formula:
Where:
- Precise Billing: Usage is calculated based on exact token counts converted to fractional pages. You are charged exactly for what you use (e.g., 320 tokens = 0.5 Pages), without rounding up to the nearest whole page.
- Cached Read: When prompt/context content has already been written to the model provider's prompt cache, subsequent calls re-using that content are billed at the Cached Read rate instead of the standard Input rate (typically a fraction of the Input price). Caching is applied automatically when supported by the provider.
- Cached Write: A premium charged the first time content is written to the cache. Currently this is only metered for Anthropic (Claude) models, where cache writes are billed at the Cached Write rate (approximately 1.25× the standard Input rate, based on the default 5-minute ephemeral cache). Other providers do not meter cache writes separately and are shown as — in the tables.
- Reasoning Tokens: Reasoning and tool calling tokens are counted as Output for the credit usage calculation.
Novis Smart & Novis Pro
Models from OpenAI, Anthropic, Google, xAI, Mistral, and other providers are available individually in the platform—organized into Smart, Pro, and Ultra tiers (see tables below).
Novis Smart and Novis Pro are routing tiers developed by Novis—not fixed single models. They adapt within the same chat based on use case, context, document length, and other signals. Routing runs in the background so Authorised Users get the best experience without choosing a model manually. A proprietary system continuously selects the top performer in each tier using the latest public benchmarks, industry-specific evaluations, Novis private benchmarks, and applied Novis fine-tuning.
- Novis Smart — Adapts in-session for everyday tasks (chat, follow-ups, summaries, light analysis).
- Novis Pro — Adapts in-session for deep reasoning, analysis, redlines, and complex document generation.
Individual third-party models in the tables below may also be selected directly, or may power Novis Smart and Novis Pro routes depending on your configuration, data zone, and availability.
Primary Models
Core models available by default (subject to availability) to customers within Novis.
Smart Models
| Model | Developer | Sub-processor | Type | Data zones | Input (C) | Cached Read (C) | Cached Write (C) | Output (C) |
|---|---|---|---|---|---|---|---|---|
| Novis Smart | Novis | Novis | Smart | US, EU, Global | 1.00 | 0.10 | — | 6.00 |
| GPT-5.4 Mini | OpenAI | OpenAI, Microsoft Azure | Smart | US, EU | 0.48 | 0.048 | — | 2.88 |
| Claude Haiku 4.5 | Anthropic | Anthropic, Google Cloud | Smart | US, EU, AP, Global | 0.64 | 0.064 | 0.8 | 3.2 |
| Gemini 3.5 Flash | Google Cloud | Smart | US, EU, Global | 0.96 | 0.096 | — | 5.76 | |
| Mistral Medium 3.1 | Mistral AI | Mistral AI | Smart | EU | 0.256 | n/a | — | 1.28 |
Pro Models
| Model | Developer | Sub-processor | Type | Data zones | Input (C) | Cached Read (C) | Cached Write (C) | Output (C) |
|---|---|---|---|---|---|---|---|---|
| Novis Pro | Novis | Novis | Pro | US, EU, Global | 2.00 | 0.20 | — | 12.00 |
| GPT-5.4 (≤ 272K T) | OpenAI | OpenAI, Microsoft Azure | Pro | US, EU | 1.6 | 0.16 | — | 9.6 |
| GPT-5.4 (> 272K T) | OpenAI | OpenAI, Microsoft Azure | Pro | US, EU | 3.2 | 0.32 | — | 14.4 |
| GPT-5.3 Codex | OpenAI | OpenAI, Microsoft Azure | Pro | US, EU | 1.12 | 0.112 | — | 8.96 |
| Claude Sonnet 4.6 | Anthropic | Anthropic, Google Cloud | Pro | US, EU | 1.92 | 0.192 | 2.4 | 9.6 |
| Gemini 3.1 Pro (≤ 200K T) | Google Cloud | Pro | US, EU, Global | 1.28 | 0.128 | — | 7.68 | |
| Gemini 3.1 Pro (> 200K T) | Google Cloud | Pro | US, EU, Global | 2.56 | 0.256 | — | 11.52 | |
| Grok 4.3 (≤ 272K T) | xAI | X.AI LLC | Pro | US, EU | 0.80 | 0.13 | — | 1.60 |
| Grok 4.3 (> 272K T) | xAI | X.AI LLC | Pro | US, EU | 1.60 | 0.26 | — | 3.20 |
Ultra Models
| Model | Developer | Sub-processor | Type | Data zones | Input (C) | Cached Read (C) | Cached Write (C) | Output (C) |
|---|---|---|---|---|---|---|---|---|
| GPT-5.5 (≤ 272K T) | OpenAI | OpenAI, Microsoft Azure | Ultra | US, EU | 3.2 | 0.32 | — | 19.2 |
| GPT-5.5 (> 272K T) | OpenAI | OpenAI, Microsoft Azure | Ultra | US, EU | 6.4 | 0.64 | — | 28.8 |
| Claude Opus 4.7 | Anthropic | Anthropic, Google Cloud | Ultra | US, EU, Global | 3.2 | 0.32 | 4.0 | 16.0 |
Secondary Open Models
Models available on customer request (Team/Enterprise plans).
| Model | Developer | Sub-processor | Type | Data zones | Input (Credit) | Cached Read (Credit) | Cached Write (Credit) | Output (Credit) |
|---|---|---|---|---|---|---|---|---|
| Mistral Large 3 | Mistral AI | Mistral AI, Google Cloud | Smart | US, EU | 0.3 | n/a | — | 1.0 |
| DeepSeek-V3.2 | DeepSeek | Google Cloud, Together Computer | Smart | US, Global | 0.4 | n/a | — | 1.1 |
| MiniMax M2 | MiniMax | Google Cloud, Nebius | Smart | US, EU, Global | 0.2 | n/a | — | 0.8 |
| Kimi-K2-Thinking | Moonshot | Google Cloud, Nebius, Together Computer | Smart | US, EU, Global | 0.4 | n/a | — | 1.6 |
| GLM-4.7 | Z.AI | Google Cloud, Nebius, Together Computer | Smart | US, EU, Global | 0.4 | n/a | — | 1.4 |
| DeepSeek-R1 (0528) | DeepSeek | Google Cloud, Together Computer, Nebius | Pro | US, EU | 0.9 | n/a | — | 3.5 |
Customers may contact support@novis.ai with questions or concerns, and sign up to receive notifications about new models.
Ulisse AI Ltd (d/b/a Novis)
71-75 Shelton Street, Covent Garden,
London, WC2H 9JQ, United Kingdom
Company number 15280517
Email: support@novis.ai
Legal notices: legal@novis.ai