Models
AI models available in Foxl
Foxl supports models from multiple providers. Use Foxl relay credits, your own API keys (BYOK), or your ChatGPT Plus/Pro subscription (OAuth).
Foxl Relay Models
Included with your Foxl plan. No setup needed - just sign in and chat.
| Model | Provider | Context | Best For |
|---|---|---|---|
| Claude Fable 5 | Bedrock | 1M tokens | State-of-the-art on most benchmarks; ambitious long-running work, advanced vision (Pro/Ultra only) |
| Claude Opus 4.8 | Bedrock | 1M tokens | New default; SOTA on SWE-bench Pro / Verified, longer autonomous runs, lower output variance for enterprise workflows |
| Claude Opus 4.7 | Bedrock | 1M tokens | Previous flagship; same API surface as 4.8 |
| Claude Opus 4.6 | Bedrock | 1M tokens | Older Opus; still 1M context |
| Claude Sonnet 4.6 | Bedrock | 1M tokens | Best balance of speed and quality |
| Claude Haiku 4.5 | Bedrock | 200K tokens | Fastest model, near-frontier intelligence |
| GLM 5 | Bedrock (Z.AI) | 200K tokens | Agentic coding, long-horizon tasks |
| Kimi K2.5 | Bedrock (Moonshot) | 256K tokens | Agentic coding and reasoning, vision |
| GPT-5.5 | Bedrock (OpenAI) | 1M tokens | OpenAI frontier reasoning, on your Foxl credits |
| GPT-5.4 | Bedrock (OpenAI) | 1M tokens | General-purpose OpenAI reasoning, on your Foxl credits |
GPT-5.5 and GPT-5.4 run on Foxl relay credits through Amazon Bedrock's native OpenAI Responses API - no ChatGPT subscription or OpenAI key required. They are also available via your own ChatGPT subscription (OAuth) below if you prefer to use that instead.
Claude Fable 5 ships with built-in safeguards. Prompts in sensitive domains (cybersecurity, biology, chemistry, health) are automatically answered by Opus 4.8 instead and billed at Opus rates. Fable 5 is a Pro/Ultra model and is not available on the free tier.
Subscription (OAuth) Models
Use your existing subscriptions. Desktop only - see Providers for setup. Foxl calls the vendor API directly with your OAuth token; tool use, adaptive thinking, and streaming all flow through Foxl's normal agent loop.
Claude Code (Anthropic Pro/Max)
| Model | Context | Best For |
|---|---|---|
| Fable 5 (Claude Code mode) | 1M tokens | State-of-the-art; most capable for ambitious, long-running work |
| Opus 4.8 (Claude Code mode) | 1M tokens | New flagship; coding, agentic workflows, long-horizon autonomy |
| Opus 4.7 (Claude Code mode) | 1M tokens | Stable Opus tier; same API surface as 4.8 |
| Sonnet 4.6 (Claude Code mode) | 200K tokens | Balanced speed and quality (200K cap on subscription pool) |
Claude Code (OAuth) runs in compatibility mode so requests route through your Claude Pro/Max subscription instead of pay-as-you-go "Extra usage". In this mode, Foxl-specific tools (memory, subagents, schedules, channel send, browser extension, view image) are disabled - only Bash, Read, Grep, and WebFetch are available. Haiku 4.5 is not exposed in this mode. For the full Foxl tool surface, use an Anthropic API key (BYOK) or the foxl.ai relay.
OpenAI (ChatGPT Plus/Pro)
| Model | Context | Best For |
|---|---|---|
| GPT-5.5 | 1M tokens | Latest frontier, reasoning + streaming |
| GPT-5.4 | 1M tokens | General-purpose reasoning |
| GPT-5.4 Mini | 400K tokens | Fast, cost-aware tasks |
| GPT-5.3 Codex | 400K tokens | Code-heavy tasks |
Also exposed as a built-in tool: gpt-image-2 via the generate_image tool. No API key, no per-image billing - powered by your ChatGPT Plus/Pro OAuth session. Accepts up to 10 input images for edit / compose mode.
Gemini CLI (Google)
| Model | Context | Best For |
|---|---|---|
| Gemini 2.5 Pro | 1M tokens | Long documents, deep analysis |
| Gemini 2.5 Flash | 1M tokens | Fast, cost-effective |
BYOK Models
Bring your own API key to access models from any provider. See AI Providers for setup.
| Model | Provider | Context | API Key From |
|---|---|---|---|
| Claude Fable 5 | Anthropic | 1M tokens | console.anthropic.com |
| Claude Opus 4.8 | Anthropic | 1M tokens | console.anthropic.com |
| Claude Opus 4.7 | Anthropic | 1M tokens | console.anthropic.com |
| Claude Opus 4.6 | Anthropic | 1M tokens | console.anthropic.com |
| Claude Sonnet 4.6 | Anthropic | 1M tokens | console.anthropic.com |
| Claude Haiku 4.5 | Anthropic | 200K tokens | console.anthropic.com |
| GPT-4.1 | OpenAI | 1M tokens | platform.openai.com |
| GPT-4.1 Mini | OpenAI | 1M tokens | platform.openai.com |
| o3 | OpenAI | 200K tokens | platform.openai.com |
| Gemini 2.5 Pro | 1M tokens | aistudio.google.com | |
| Gemini 2.5 Flash | 1M tokens | aistudio.google.com | |
| Llama 3, Mistral, etc. | Ollama | Varies | Free - ollama.com |
Adaptive Thinking
Claude Fable 5, Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6 support adaptive thinking - Claude dynamically decides when and how much to think based on the complexity of your request. No manual budget setting needed.
- Simple questions: Claude responds directly without thinking overhead
- Complex problems: Claude automatically engages deep reasoning
- Agentic workflows: Claude can think between tool calls (interleaved thinking)
Adaptive thinking (type: "adaptive") is the recommended mode for Fable 5, Opus 4.8 / 4.7 / 4.6, and Sonnet 4.6 (Fable 5 accepts only adaptive). Haiku 4.5 uses type: "enabled" with a budget_tokens parameter instead.
Opus 4.7 and 4.8 also reject non-default temperature / top_p / top_k with HTTP 400 - Foxl strips those automatically on those models, so the API uses its calibrated default.
You can toggle thinking on/off in the model selector. For adaptive-thinking models, Settings also exposes a reasoning effort level (the options come from the selected model's capabilities - up to low / medium / high / xhigh / max); models without adjustable reasoning show no effort selector.
Task Budgets (Beta, Fable 5 and Opus 4.7+)
Settings includes a Task Budget selector (off / 25K / 50K / 100K / 250K / 500K tokens). When set, the model receives an output_config.task_budget and the task-budgets-2026-03-13 beta header on every request, and uses the remaining budget to plan and pace its work across one agentic turn (tool calls + thinking + response).
- Advisory, not enforced. The model treats it as guidance, not a hard ceiling.
- Minimum 20K when enabled. Anthropic rejects positive values below that.
- Plumbs through the foxl relay path and the direct Anthropic API path. Bedrock Converse rejects
output_config.task_budget("Extra inputs are not permitted"), so Foxl omits it on the direct-Bedrock route - the budget applies only on relay and Anthropic-API transports.
Thinking and Cost
Thinking consumes output tokens. When thinking is enabled, the model may use 2-10x more output tokens depending on task complexity. This directly increases credit cost:
- Simple question without thinking: ~0.01 credits (Sonnet)
- Same question with thinking: ~0.05-0.15 credits (Sonnet)
- Complex reasoning with thinking: ~0.30-1.0 credits (Opus)
For cost-sensitive usage, disable thinking for simple tasks. For complex coding, analysis, or multi-step reasoning, thinking significantly improves quality and is worth the extra cost. See Credits for detailed per-token pricing.
Model Selection
Desktop App
Click the model name in the chat input area to switch models. Your selection persists across conversations.
Web App
Click the model selector dropdown to choose a different model. The selection is saved in your browser.
Bring Your Own Key (Desktop Only)
In the desktop app, you can use your own API keys instead of Foxl credits:
- Go to Settings
- Select a provider (Anthropic, OpenAI, Google, etc.)
- Enter your API key
- Select a model from that provider
When using your own keys, no Foxl credits are consumed.
Subscription OAuth
You can use your ChatGPT Plus/Pro subscription directly - no API key needed.
OpenAI OAuth
- Run
npx @openai/codex loginto create~/.codex/auth.json - Select OpenAI (OAuth) in Settings > Providers - Foxl calls OpenAI's Codex Responses API directly with your subscription token
Foxl auto-detects your credentials. See AI Providers for details on all supported providers.