Pricing
Pay for the sandbox.
Compute is the thing we can buy well, so compute is what we sell. Tokens carry a small markup you can opt out of, allotments reset instead of expiring, and prices follow the market you actually live in.
Free
Enough to find out whether you'll actually ship from your phone.
Free forever
- 30 minutes of agent time per month
- 1 connected repo
- Bring your own model key for unlimited tokens
- Full diff review and PR creation
Starter
For the side project you keep thinking about on the train.
Billed monthly, cancel anytime
- 6 hours of agent time per month
- $2.50 of model credits included
- 3 connected repos, 2 runs at once
- 2 vCPU / 4 GB sandboxes
- Unlimited tokens with your own API key
Pro
Your primary development environment, which happens to be a phone.
Billed monthly, cancel anytime
- 25 hours of agent time per month
- $6.50 of model credits included
- Live preview URLs for running apps
- 4 vCPU / 8 GB sandboxes, 5 runs at once
- Frontier models: Sonnet 4.5, GPT-5, Gemini 2.5 Pro
Team
Shared repos, pooled usage, one invoice.
Billed monthly, cancel anytime
- Everything in Pro, per seat
- Pooled usage across the team
- Shared workspaces and run history
- SSO and audit log
- Minimum 3 seats
Plan allotments reset each billing period and do not roll over. Separately purchased top-up credits never expire — we do not run an expiring-credit business.
Agent time
What the allotment actually buys.
Agent time is billed on wall-clock seconds your sandbox is awake. These are real averages from the beta, not marketing numbers.
| Plan | Small fix ~2 min | Feature ~6 min | Refactor ~13 min |
|---|---|---|---|
| free | 20 runs | 5 runs | 2 runs |
| starter | 240 runs | 65 runs | 27 runs |
| pro | 1000 runs | 272 runs | 115 runs |
A “small fix” is a one-file change with a test run. A “refactor” touches a dozen files and runs the full suite twice.
Models
Every model, at close to cost.
Credits are charged at the provider's list price plus a small routing markup — 15% on Google models, 25% on everything we buy through a reseller. Attach your own key and that markup disappears.
| Model | Input / 1M | Output / 1M | Context | Free plan |
|---|---|---|---|---|
Gemini 2.5 Flash Default. Fast enough to feel live on mobile, cheap enough to include free. | $0.30 | $2.50 | 1049k | Included |
Gemini 2.5 Pro Large-context reasoning. Best for multi-file refactors and debugging. | $1.25 | $10.00 | 1049k | Paid plans |
Claude Sonnet 4.5 Strongest agentic coding. Premium route — the most expensive tokens on the platform. | $3.00 | $15.00 | 200k | Paid plans |
Claude Haiku 4.5 Most of Sonnet's tool-use discipline at a third of the price. | $1.00 | $5.00 | 200k | Paid plans |
GPT-5 Strong planning. Good second opinion when a run stalls. | $1.25 | $10.00 | 400k | Paid plans |
DeepSeek V3.2 Cheapest capable coder. Carries the free tier in price-sensitive markets. | $0.28 | $1.10 | 128k | Included |
Qwen3 Coder Open weights. The hedge against any single vendor repricing us. | $0.30 | $1.20 | 262k | Included |
Gemini 2.5 Flash-Lite The cheapest thing that can still follow a tool schema. Good for bots that run often. | $0.10 | $0.40 | 1049k | Included |
Claude Opus 4.1 For the refactor nothing else gets right. Five times Sonnet's price; use it on purpose. | $15.00 | $75.00 | 200k | Paid plans |
GPT-5 Codex GPT-5 tuned for long agentic coding sessions; paces its own thinking. | $1.25 | $10.00 | 400k | Paid plans |
GPT-5 mini Most of GPT-5's tool discipline at a fifth of the price. | $0.25 | $2.00 | 400k | Included |
gpt-oss 120B OpenAI's open weights. Blisteringly fast on Groq and Cerebras. | $0.10 | $0.50 | 131k | Included |
Grok Code Fast xAI's coding model: quick, cheap, keen on tool calls. | $0.20 | $1.50 | 256k | Included |
Grok 4 xAI's frontier reasoner. | $3.00 | $15.00 | 256k | Paid plans |
Kimi K2 Trillion-parameter open MoE with excellent tool use. Very fast on Groq. | $0.60 | $2.50 | 262k | Included |
GLM-4.6 Z.ai's open coder; close to Sonnet on agentic benchmarks at a tenth of the price. | $0.60 | $2.20 | 200k | Included |
MiniMax M2 Small active footprint, built for agents and coding loops. | $0.30 | $1.20 | 205k | Included |
Devstral Medium Mistral's agentic coder. European option that is not a compromise. | $0.40 | $2.00 | 131k | Included |
Mistral Medium 3.1 General-purpose Mistral, good at long documents. | $0.40 | $2.00 | 131k | Included |
Llama 4 Maverick Meta's open MoE with a million-token window. | $0.15 | $0.60 | 1049k | Included |
List prices as of 2026-09-01. When a provider cuts prices, your credits stretch further the same day — we do not hold the old rate.
Already paying Anthropic or OpenAI?
Paste the key once. It is encrypted with a per-account envelope key, decrypted only inside the sandbox running your job, and never written to logs. Tokens bill to your provider account and touch none of your Pocket credits.
You still use agent time, which is what your plan is actually for. Available on every plan, free included.
Or skip our sandboxes entirely.
Attach a GCP instance, an AWS box, a Vast.ai GPU, or the desktop sitting at home. Pocket dispatches the run over a tunnel, and the agent-time meter does not move.
It is an odd thing to offer, since it removes our main revenue line. We would rather be the interface you keep than the bill you resent.
Top-ups
Credits that don't expire.
The standard playbook here is to sell prepaid credits with an expiry date and book the unused balance as profit. We are not doing that, for two reasons.
The first is that it only makes money when customers stop using the product, which is a strange thing to optimise for. The second is that expiring stored value attracts unclaimed-property law in several of the markets we want to be in.
Plan allotments reset monthly. Money you hand us separately stays yours until you spend it.
$5
$5.00 in credits
$20
$21.00 in credits
+$1 bonus
$50
$55.00 in credits
+$5 bonus
Thirty minutes, no card.
Enough to connect a repo and find out whether you actually ship from a phone.
Start free