# Models overview

> Compare the AI models available in Specify, including plan availability, context size, and cost.

Specify offers six models from the Anthropic Claude and OpenAI GPT families. Choose a model for each conversation in the chat message box, and set the default models for automations and document generation in workspace settings.

<Screenshot name="model-picker" alt="The model menu in the chat message box listing Claude Opus 4.6, Claude Sonnet 4.6, Claude Haiku 4.5, GPT-6 Sol, GPT-5.6 Terra, and GPT-6 Luna" />

## Compare models

| Model | Provider | Available on | Context | Max output | Input / output price (per 1M tokens) |
| --- | --- | --- | --- | --- | --- |
| **Claude Opus 4.6** | Anthropic | Pro and above | 1M | 128K | $5.00 / $25.00 |
| **Claude Sonnet 4.6** | Anthropic | Pro and above | 1M | 64K | $3.00 / $15.00 |
| **Claude Haiku 4.5** | Anthropic | All plans, including Free | 200K | 64K | $1.00 / $5.00 |
| **GPT-6 Sol** | OpenAI | Pro and above | 1.05M | 128K | $2.00 / $10.00 |
| **GPT-5.6 Terra** | OpenAI | Pro and above | 1.05M | 128K | $2.00 / $12.00 |
| **GPT-6 Luna** | OpenAI | All plans, including Free | 1.05M | 128K | $0.10 / $0.50 |

- Prices are the providers' standard rates (checked September 2026). Specify deducts the actual cost calculated at these rates from your workspace AI budget. For details, see [Usage and AI budget](/en/models/usage).
- Every model supports tool calls (searching knowledge, reading and writing documents, and more) and parallel tool calls.
- The default model for new conversations is **GPT-6 Luna**.

## Model characteristics

<CardGroup cols={2}>
  <Card icon="sparkles" title="GPT-6 Luna: default model">The least expensive model. Try it first for everyday questions, search, and summaries. It's also available on the Free plan.</Card>
  <Card icon="search" title="Claude Haiku 4.5: lightweight Claude">The least expensive Claude model. It's good at looking up information and summarizing, but weaker at multi-step document editing.</Card>
  <Card icon="file" title="Claude Sonnet 4.6: document work">Reliable for writing and editing documents, following skills, and connector tasks. Supports a 1M-token context.</Card>
  <Card icon="library" title="Claude Opus 4.6: deepest reasoning">Use it for the hardest analysis and for designing and verifying the structure of long documents. It's the most expensive.</Card>
  <Card icon="blocks" title="GPT-6 Sol: general high performance">A high-performance OpenAI model that costs less than Sonnet. Use it when you want to try a complex request with a model from a different family.</Card>
  <Card icon="workflow" title="GPT-5.6 Terra: OpenAI alternative">An OpenAI model with the same input price as Sol and a slightly higher output price. Use it to compare results with other OpenAI models.</Card>
</CardGroup>

For which model to use for each task, see [Choosing a model](/en/models/choosing).

## Reasoning effort

Every model lets you set **reasoning effort** to one of six levels: **Off**, **Low**, **Medium**, **High**, **Extra**, or **Max**. Higher effort means deeper reasoning before answering, but responses are slower and use more tokens.

| Level | Minimum tokens used for reasoning |
| --- | --- |
| Off | No reasoning |
| Low | About 1,000 |
| Medium | About 4,000 |
| High | About 16,000 |
| Extra | About 32,000 |
| Max | About 60,000 |

<Note>
  On the Free plan, reasoning effort is fixed at **Off**. When you switch models, the current reasoning effort is adjusted to the range the new model supports.
</Note>

## Where to choose a model

| Location | Applies to |
| --- | --- |
| The model button in the chat message box | The whole current conversation. It can't be changed while a response is in progress. |
| **Settings** → **Workspace general** → **Default AI Model** | Automations in this workspace |
| **Settings** → **Workspace general** → **Doc Generation Model** | Document automation (project creation and automatic updates) |

Models that aren't available on your plan show a lock in the menu, and selecting one shows a plan upgrade prompt. Specify never silently swaps the model you requested for a different one.