Background
Updated on May 15, 2026
Scopes of Data
Scope of data decides which saved context is used for a chat. Every message you send is routed into either global scope or project scope, and that choice determines which memories and files the system can see. This keeps everyday chats, project chats, and persona chats cleanly separated so replies stay relevant and nothing leaks across boundaries you didn't intend.
Understanding scopes helps you organize your work: put sensitive or project-specific context in a project, keep general preferences in global, and use personas when you want consistent voice across both without merging the underlying data.
Global
Global scope is used when you chat outside any project. It draws from your account-level memories and account-level files—the context you've chosen to share across all everyday conversations.
Global scope does not pull in project-only details. If you have a project for work notes or a side venture, those memories and files stay out of your regular chats. This helps keep everyday conversations separate from focused project work, so you can switch contexts without worrying about cross-talk.
Project
Project scope is used when you chat inside a specific project. It uses that project's memories and files only—nothing from other projects or from your global account.
Each project stays isolated. Details from one project do not spill into other projects or into regular global chats. That isolation is intentional: you can have a project for a client, another for personal research, and another for a hobby, and each one keeps its own context clean.
Persona
Persona is a style layer that can be used in both global and project chats. It keeps voice, tone, and behavior consistent across spaces without merging the underlying memory or file stores.
Persona can travel across chats, while global and project memory spaces remain separate. You get continuity in tone—the same personality whether you're in a global chat or a project chat—without mixing private context. It's the best of both: consistent voice, strict data boundaries.
Files
Files follow the same scope rules as memories. Global files are available in global chats; project files are available only when you're chatting inside that project. There is no shared file pool that crosses scope boundaries.
Upload files where they belong: account-level docs for everyday use, project-specific docs inside the project. The system retrieves only the files that match the current scope when you send a message.
Customization And Scope Interaction
Customization settings are account-level and can be managed from Account in the Customization and Memories sections. The same options are also available in the chat account modal.
In regular chats, customization settings guide reply style—how the assistant structures responses, how formal or casual it sounds, and other preferences. When a persona is active, persona guidance takes priority so tone stays consistent for that persona across every chat.
Global chats continue to use global memory space, and project chats continue to use project memory space. Persona can move across both without merging those spaces. You get a single, coherent voice without blurring the lines between work and personal context.
Models
Model providers have different strengths. Use the summaries below to choose a provider, then pick a model family that matches the task.
OpenAI (GPT)
Strong all-around for writing, planning, coding, and everyday questions. Includes GPT models from OpenAI plus GPT OSS variants served through Fireworks and Cerebras.
| Model | Requests | Best For |
|---|---|---|
| GPT-4o · May 2024 | 2 premium requests | May 2024 checkpoint of gpt-4o for that special voice. |
| GPT-4o · Aug. 2024 | 1 premium request | August 2024 checkpoint of gpt-4o with enhanced capabilities. |
| GPT-4o · Nov. 2024 | 1 premium request | November 2024 checkpoint of gpt-4o with latest improvements. |
| GPT-4o | 1 premium request | OpenAI's default gpt-4o through API. |
| GPT-4o mini | 1 base request | Agile, cost-efficient 4o variant ideal for everyday conversation. |
| Model | Requests | Best For |
|---|---|---|
| GPT-3.5 Turbo | 1 base request | Legacy GPT model for cheaper chat and non-chat tasks. |
| GPT-3.5 Turbo · 0125 | 1 base request | Pinned January 2024 snapshot of GPT-3.5 Turbo. |
| GPT-3.5 Turbo · 1106 | 1 base request | Pinned November 2023 snapshot of GPT-3.5 Turbo. |
| Model | Requests | Best For |
|---|---|---|
| GPT audio mini | 2 base requests | Cost-efficient audio-native chat model. Supports text + audio output in chat completions. |
| GPT audio mini · Oct. 2025 | 2 base requests | Pinned October 2025 snapshot of GPT Audio mini for stable behavior. |
| GPT audio mini · Dec. 2025 | 2 base requests | Pinned December 2025 snapshot of GPT Audio mini for stable behavior. |
| Model | Requests | Best For |
|---|---|---|
| GPT-5.4 | 2 premium requests | Latest GPT-5.4 flagship chat model with stronger reasoning and accuracy. |
| GPT-5.4 mini | 1 premium request | Higher-capability GPT-5.4 mini for high-volume coding, computer use, and subagent workflows. |
| GPT-5.4 nano | 1 base request | Cheapest GPT-5.4-class model for simple high-volume tasks such as extraction, ranking, and lightweight subagents. |
| Model | Requests | Best For |
|---|---|---|
| GPT-5.3-Codex | 1 premium request | The most capable agentic coding model to date. Optimized for agentic coding tasks in Codex or similar environments. 400K context, 128K max output. Reasoning off by default. |
| GPT-5.3 latest | 1 premium request | GPT-5.3 model used in ChatGPT. Best general-purpose model with high intelligence and vision support. Pricing assumed same as 5.2/5.1 chat latest until announced. |
| Model | Requests | Best For |
|---|---|---|
| GPT-5.2 latest | 1 premium request | GPT-5.2 model used in ChatGPT. Best general-purpose model with high intelligence and vision support. |
| GPT-5.2 | 1 premium request | Pinned GPT-5.2 snapshot for stable behavior. |
| Model | Requests | Best For |
|---|---|---|
| GPT-5.1 | 1 premium request | Pinned snapshot gpt-5.1-2025-11-13. The most intelligent model yet, with faster responses and increased steerability. |
| GPT-5.1 latest | 1 premium request | GPT-5.1 model used in ChatGPT. Continuously updated for the latest chat improvements. |
| GPT-5.1 codex | 1 premium request | GPT-5.1 optimized for agentic coding in Codex. 400K context, 128K max output. |
| GPT-5.1 codex mini | 1 base request | Smaller, more cost-effective, less-capable version of GPT-5.1-Codex. 400K context, 128K max output. |
| Model | Requests | Best For |
|---|---|---|
| GPT-5 | 1 premium request | Frontier reasoning depth with best-in-class reliability. |
| GPT-5 codex | 1 premium request | Enhanced code reasoning while staying conversation-friendly. |
| GPT-5 latest | 1 premium request | Continuously tuned GPT-5 chat experience with the latest guardrails. |
| GPT-5 mini | 1 base request | Responsive, budget-friendly member of the GPT-5 family. |
| GPT-5 nano | 1 base request | Ultra light-touch assistant for simple interactions. |
| Model | Requests | Best For |
|---|---|---|
| GPT-4.1 | 1 premium request | GPT-4 refinement designed for coding with broad tool compatibility. |
| GPT-4.1 mini | 1 base request | Compact GPT-4.1 option for consistent tone and speed. |
| GPT-4.1 nano | 1 base request | Minimal footprint 4.1 for background automation tasks. |
| Model | Requests | Best For |
|---|---|---|
| o3 | 1 premium request | Reasoning-focused o-series model optimized for long horizon tasks. |
| o4 mini | 1 premium request | Lean o-series model for high volume creative projects. |
| o3 mini | 1 premium request | Balanced o-series variant with emphasis on tool use during reasoning. |
| Model | Requests | Best For |
|---|---|---|
| GPT OSS 120b | 1 base request | OpenAI's open-weight 117B MoE via Fireworks. Production-grade reasoning, agentic tasks, function calling. 131k context. Does not support web search or image input. |
| GPT OSS 120B Fast | 1 premium request | OpenAI's GPT OSS 120B routed through Cerebras chat completions for very fast tool-capable replies. 131k context. Does not support web search or image input. |
| GPT OSS 20b | 1 base request | OpenAI's open-weight 21B MoE via Fireworks. Lower latency, local or specialized use-cases. 131k context. Does not support web search, image input, or function calling. |
Google (Gemini)
Great for long instructions, large context, and quick iteration on bigger tasks.
| Model | Requests | Best For |
|---|---|---|
| Gemini 2.5 Pro | 1 premium request | state-of-the-art multipurpose model, which excels at coding and complex reasoning tasks. |
| Gemini 2.5 Flash | 1 base request | first hybrid reasoning model which supports a 1M token context window and has thinking budgets. |
| Gemini 2.5 Flash-Lite | 1 base request | smallest and most cost effective model, built for at scale usage. |
| Model | Requests | Best For |
|---|---|---|
| Gemini 3.1 Pro Preview | 3 premium requests | Next iteration of Gemini 3 Pro: performance, behavior, and intelligence improvements. 1M/64k context. Agentic workflows, autonomous coding, complex multimodal. Jan 2025. |
| Gemini 3.1 Flash-Lite Preview | 1 base request | Most cost-efficient Gemini 3.1 model for high-volume agentic tasks, translation, and simple data processing. 1M/65k context. Preview; may have stricter rate limits. |
| Model | Requests | Best For |
|---|---|---|
| Gemini 3 Flash Preview | 1 base request | Preview of Gemini 3 Flash. 1M/64k context. Jan 2025. |
Anthropic (Claude)
Good for careful writing, nuanced edits, and thoughtful longer responses.
| Model | Requests | Best For |
|---|---|---|
| Claude Opus 4.6 | 5 premium requests | Anthropic's most advanced Claude model. Exceptional emotional intelligence and warmth paired with adaptive thinking that scales to the complexity of your request. |
| Claude Opus 4.5 | 5 premium requests | Previous flagship Claude with strong reasoning capabilities. Great balance of intelligence and accessibility. |
| Claude Sonnet 4.6 | 5 premium requests | Anthropic's most capable Sonnet yet. Full upgrade across coding, long-context reasoning, agent planning, and design. 1M token context window in beta. Same pricing as Sonnet 4.5. |
| Claude Sonnet 4.5 | 5 premium requests | Anthropic's balanced Claude model with strong reasoning and efficiency. |
| Claude Haiku 4.5 | 1 premium request | Anthropic's fastest Claude model, optimized for speed and cost efficiency. |
xAI (Grok)
Good for quick back-and-forth, practical answers, and fast drafting.
| Model | Requests | Best For |
|---|---|---|
| Grok-3 Mini | 1 base request | Compact Grok-3 variant for cost-effective conversations. |
| Model | Requests | Best For |
|---|---|---|
| Grok-4.20 Reasoning | 1 premium request | xAI's flagship Grok 4.20 reasoning model with a 2M-token context window, stronger multi-step reasoning, and native tool support. |
| Grok-4.20 Non-Reasoning | 1 premium request | Latency-optimized Grok 4.20 variant with a 2M-token context window, image understanding, and native tool support. |
| Model | Requests | Best For |
|---|---|---|
| Grok 4.3 | 1 premium request | xAI's latest flagship Grok model with 1M context, image input, configurable reasoning, and strong agentic tool use. |
DeepSeek
Strong for reasoning and complex tasks. DeepSeek v3.x models are available on all tiers; DeepSeek V4 Pro is Premium with 1M context and function calling. No web search or images.
| Model | Requests | Best For |
|---|---|---|
| DeepSeek V4 Pro | 1 premium request | DeepSeek-V4-Pro via Fireworks: flagship open-source 1.6T MoE model for frontier reasoning, advanced coding, and long-context agentic workflows. 1M context. Function calling supported. Uses 1 premium request per send before length multipliers. Does not support web search or image input. |
Qwen
Strong for multimodal chat, tool use, and general flagship work. Available on all tiers through Fireworks serverless. Supports image input, but not web search.
| Model | Requests | Best For |
|---|---|---|
| Qwen 3.6 Plus | 2 base requests | Alibaba's flagship closed Qwen model via Fireworks. 396B MoE with function calling and image input support. Available serverless on just4o.chat at the base tier. Uses 2 base requests per send before length multipliers. Does not support web search. |
Moonshot (Kimi)
Good for complex reasoning, multimodal agentic tasks, and long-horizon coding. Kimi K2.5 and K2.6 support images. No web search.
| Model | Requests | Best For |
|---|---|---|
| Kimi K2.5 | 1 premium request | Moonshot AI's flagship agentic model via Fireworks. Unifies vision and text, thinking and non-thinking. 262k context. Supports image input. Does not support web search. |
| Kimi K2.6 | 2 premium requests | Moonshot AI's Kimi K2.6 via Fireworks: open-source, native multimodal agentic model for long-horizon coding, coding-driven design, autonomous execution, and task orchestration. 1T MoE, 262k context. Supports image input and function calling. Uses 2 premium requests per send before length multipliers. Does not support web search. |
MiniMax
Strong for coding, complex tasks, and office work. MiniMax M2.7 supports image input. Available on all tiers. No web search.
| Model | Requests | Best For |
|---|---|---|
| MiniMax M2.5 | 1 base request | MiniMax M2.5 via Fireworks: state-of-the-art coding, agentic tool use, search, and office work. 228B MoE, 196k context. Function calling supported. Does not support web search or image input. |
| MiniMax M2.7 | 1 base request | MiniMax M2.7 via Fireworks: 228B MoE model for complex agent harnesses, productivity tasks, Agent Teams, Skills, and dynamic tool search. 196k context. Supports image input and function calling. Does not support web search. |
Z.ai (GLM)
Strong for coding, reasoning, and long-horizon agentic workflows. GLM models are available through Fireworks and Cerebras; GLM 4.7 Fast is the Cerebras-backed Premium OSS variant. No web search or images.
| Model | Requests | Best For |
|---|---|---|
| GLM 4.7 Fast | 2 premium requests | Z.ai's GLM-4.7 routed through Cerebras chat completions for lower-latency coding and agentic work. 131k context. Does not support web search or image input. Cerebras currently lists it as a preview model. |
| Model | Requests | Best For |
|---|---|---|
| GLM 5.1 | 2 premium requests | Z.ai's GLM-5.1 via Fireworks: next-generation flagship for agentic engineering, stronger coding, and sustained long-horizon task performance. 202k context. Uses 2 premium requests per send before length multipliers. Does not support web search, image input, or function calling. |