# Pi

Canonical page: https://agentsky.dev/agents/pi

Fast, steerable coding agent for any compatible model.

Pi is a minimal, fast coding agent designed for interactive sessions with live steering — you can redirect or correct a running task mid-execution without restarting it. It supports a broad model set across OpenAI, DeepSeek, Google, Zhipu, and Moonshot, making it flexible for cost and language preference.

## Facts

- **Agent type:** `pi`
- **Default model:** GPT-5.6 Sol (`gpt-5.6-sol`)
- **Start in the browser:** https://agentsky.dev/playground/enter?agentType=pi&llm=gpt-5.6-sol&source=agent-md
- **API reference:** https://agentsky.dev/docs
- **Billing:** Running sessions are billed by compute time and model tokens; parked agents incur no compute charge.

## Models it runs on

USD per 1M tokens.

| Model | Id | Input | Output |
|---|---|---|---|
| GPT-5.6 Sol (default) | `gpt-5.6-sol` | $3.50 | $21.00 |
| GPT-5.6 Terra | `gpt-5.6-terra` | $1.40 | $8.40 |
| GPT-5.6 Luna | `gpt-5.6-luna` | $0.14 | $0.84 |
| GPT-6 Astra | `gpt-6-astra` | $7.00 | $35.00 |
| GPT-6.1 Sol | `gpt-6.1-sol` | $1.40 | $7.00 |
| GPT-6 Sol | `gpt-6-sol` | $1.40 | $7.00 |
| GPT-6 Luna | `gpt-6-luna` | $0.07 | $0.35 |
| Gemini 4 Argon | `gemini-4-argon` | $4.40 | $22.00 |
| Gemini 3.5 Flash-Lite | `gemini-3.5-flash-lite` | $0.33 | $2.75 |
| Gemini 3.8 Flash | `gemini-3.8-flash` | $1.65 | $8.25 |
| DeepSeek V4 Pro | `deepseek-v4-pro` | $0.924 | $2.772 |
| DeepSeek V4.1 Flash | `deepseek-flash` | $0.21 | $0.84 |
| GLM-5.3 | `glm-5.3` | $1.54 | $4.84 |
| GLM-5.3-Flash | `glm-5.3-flash` | $0.165 | $0.55 |
| Kimi K3.1 Pro | `kimi-k3.1-pro` | $2.20 | $11.00 |
| Kimi K3 | `kimi-k3` | $3.30 | $16.50 |
| Kimi K2.7 Code | `kimi-k2.7-code` | $1.045 | $4.40 |
| Grok 4.6 | `grok-4.6` | $2.20 | $6.60 |
| Muse Spark 1.1 | `muse-spark-1.1` | $1.375 | $4.675 |
| Muse Wonder 1.0 | `muse-wonder-1.0` | $4.40 | $22.00 |
| Muse Spark 1.3 | `muse-spark-1.3` | $1.375 | $4.675 |
| Muse Spark 1.2 | `muse-spark-1.2` | $1.375 | $4.675 |
| Claude Opus 5.5 | `claude-opus-5-5` | $2.80 | $14.00 |
| Claude Opus 5 | `claude-opus-5` | $3.50 | $17.50 |
| Claude Fable 5.1 | `claude-fable-5-1` | $7.00 | $35.00 |
| Claude Sonnet 5 | `claude-sonnet-5` | $1.40 | $7.00 |
| Claude Haiku 4.5 | `claude-haiku-4-5` | $0.70 | $3.50 |
| Claude Sonnet 5.5 | `claude-sonnet-5-5` | $1.40 | $7.00 |
| Claude Haiku 5.5 | `claude-haiku-5-5` | $0.07 | $0.35 |
| Grok 4.7 | `grok-4.7` | $2.20 | $6.60 |
| MiniMax-M3 | `minimax-m3` | $0.33 | $1.32 |
| MiniMax-M3.1-Flash | `minimax-m3.1-flash` | $0.165 | $0.66 |

## What it can do

- **Conversational AI:** Maintains coherent, multi-turn conversations at task or topic depth.
- **Research & analysis:** Reads sources and synthesises a grounded, cited answer.
- **Content creation:** Drafts, edits, and refines text for any audience or tone.
- **Cross-model support:** Run Pi with Claude, GPT, DeepSeek, Gemini, or Kimi.

## Questions

### How do I run Pi as a cloud coding agent on AgentSky without managing my own compute environment or containers?

Sign up for AgentSky and start a Pi agent via the API or web UI. No server setup required: AgentSky provisions a persistent cloud computer with Pi installed. Specify the model in the same call. Pi is lightweight and boots quickly, so initial runs start faster than heavier harnesses.

### Which models does Pi support on AgentSky, and does it work with DeepSeek, Gemini, and open-source model providers?

Pi works with GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna, Gemini 4 Argon, Gemini 3.5 Flash-Lite, Gemini 3.8 Flash, DeepSeek V4 Pro, DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3-Flash, Kimi K3.1 Pro, Kimi K3, Kimi K2.7 Code, Grok 4.6, Muse Spark 1.1, Muse Wonder 1.0, Muse Spark 1.3, Muse Spark 1.2, Claude Opus 5.5, Claude Opus 5, Claude Fable 5.1, Claude Sonnet 5, Claude Haiku 4.5, Claude Sonnet 5.5, Claude Haiku 5.5, Grok 4.7, MiniMax-M3, and MiniMax-M3.1-Flash. The model is a parameter in the API call that starts the agent, so switching providers is a one-word change with no integration rewrite.

### What does it cost to run Pi on AgentSky per task, and how is it billed?

Pi is a lightweight harness, so per-task compute costs are typically lower than heavier agents. Model costs depend on the provider and token usage. Playground coverage for Pi is expanding, so task-level median cost benchmarks are not yet published. Current per-model rates are on the pricing page.

### My Pi coding agent loses context partway through a task or stops responding without completing the requested changes — how do I debug this?

Check the task log for the last tool call or message before the stop. Pi can stall if a shell command exits with a non-zero status that Pi treats as terminal, or if the working context grows beyond the selected model's window. Keep tasks scoped to one module or file set, and use a model with a larger context window when the task reads many files simultaneously.

### If my Pi agent is mid-task and the cloud machine AgentSky runs it on gets restarted or replaced, will it lose the files it was editing?

The session workspace persists across ordinary restarts. If the machine is replaced, AgentSky restores the latest durable recovery point, preserving the working directory and recent file state. Work in progress at the exact moment of a hard replacement may replay from the last checkpoint. Because Pi is a minimal harness, replays are typically faster than heavier agents that must re-read large context windows.

### How does running Pi on AgentSky compare to running it locally on my development machine or self-hosting it on a server?

Locally, Pi stops when your laptop closes, the network drops, or the shell exits. On AgentSky it runs in a persistent cloud computer that stays alive, auto-recovers from crashes, and is reachable via API or channels like Slack and CLI without port forwarding.