Opencode for free: which models are available and what the limits are

opencode itself is free, and its built-in provider offers a set of free models including Big Pickle. Here is what those models are, how they treat your data, and what an hour of agent work costs once you move to paid models.

7 min · · opencode · models · экономия

Yes, opencode can be used for free. The agent is MIT-licensed, and its built-in provider offers several free models, including Big Pickle. There are three catches. The free models are free "for a limited time" with no published end dates or rate limits. Some of them may use your requests to improve the model. And they only work inside OpenCode itself. Below is the list as of 29.10.2026, the terms for each model, and options for when free is not enough.

What is free in opencode

OpenCode itself costs nothing. According to the download page as of 29.10.2026, it is MIT-licensed, and the only costs come from the model provider. For an overview of the agent, see what opencode is.

Free models come through the built-in opencode provider. That is OpenCode Console, a pay-as-you-go service formerly called OpenCode Zen. Some models in its pricing table are marked Free.

There are also local models and GitHub Copilot, covered below.

Free model list

Per the Console models page as of 29.10.2026, these models are free. Data terms are listed to the extent the page describes them.

Model IDKnown terms
big-picklestealth model, real name not disclosed; data may be used to improve the model
glm-5.3-flash-freeprovider states it does not store or train on data
deepseek-v4.1-flash-freeprovider states it does not store or train on data
beam-freeunavailable in Russia, Belarus and several other countries
flint-40b-freedata retained for 30 days and may be used to improve the model
mimo-v2.6-flash-freecheck terms on the models page
nemotron-3-ultra-freefor trial only; do not send personal or confidential data
nemotron-3.5-lightning-freefor trial only; same restrictions
nemotron-3.5-super-vl-freefor trial only; same restrictions

The list changes, so check the source before committing to a model.

Big Pickle: what is known

Big Pickle is a stealth model whose real name has not been disclosed. While it is free, collected data may be used to improve it.

That is the extent of public information. The underlying model, its limits, and the end of the free period are unknown. The only way to judge its quality is on your own tasks.

In practice, Big Pickle fits learning projects, experiments, and open-source code. Proprietary or NDA-covered code is better kept out of it.

Limits of the free models

Temporary by design

All free models are free "for a limited time". The docs give no end dates and no request limits. A model can disappear or become paid, so building a long-term workflow on one is risky.

Data

The opencode site says user data is stored only when free models are used. Paid requests are handled differently in that respect.

Terms vary by model. GLM-5.3-Flash Free and DeepSeek V4.1 Flash Free are stated to have no storage or training. Flint 40B Free keeps data for 30 days. Nemotron models are for trial use only. Big Pickle may use data for improvement.

Inside OpenCode only

Free chat models work only in OpenCode itself. Calling Console from another client over the API requires paid models or a Go subscription.

Regional availability

Beam Free is unavailable in Russia, Belarus and a number of other countries. For the other models, check availability in your region on the models page.

How to pick a free model

In the OpenCode UI, use the /models command. In the config, use the opencode/<model-id> format:

bash
{
  "model": "opencode/big-pickle"
}

To hide free models from the list, set an environment variable, as described on the models docs page as of 29.10.2026:

bash
export OPENCODE_DISABLE_FREE_MODELS=1

For code you would rather not share, the sensible starting points in the free set are the models with stated no-storage terms: glm-5.3-flash-free and deepseek-v4.1-flash-free.

Other free options

The providers docs as of 29.10.2026 list two more no-cost routes.

  • Local models. OpenCode detects Ollama at http://127.0.0.1:11434 automatically. LM Studio and vLLM are supported too. Requests stay on your machine, but quality and speed depend on your hardware.
  • GitHub Copilot, including the Copilot Free plan, connected via /connect.

For a broader look at free coding options, see free AI for programming.

When free stops being enough

Usually for one of three reasons: the model disappears, its data terms do not suit work code, or you need a specific Claude- or GPT-class model. There are three paid routes.

  • Paid models in OpenCode Console (formerly Zen), pay as you go.
  • An OpenCode Go subscription at $10 per month, per the Go page as of 29.10.2026.
  • Your own provider or an API gateway with a key.

How to weigh these is covered in OpenCode Zen or your own provider.

Connecting Zukko

Zukko is a pay-as-you-go gateway to model APIs. Per the Zukko docs, Claude models are added through the anthropic provider in ~/.config/opencode/opencode.json:

bash
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "anthropic": {
      "options": {
        "baseURL": "https://api.zukko.pro/v1",
        "apiKey": "fetch_..."
      },
      "models": {
        "claude-opus-5-5": {},
        "claude-sonnet-5": {}
      }
    }
  }
}

The base URL includes the /v1 suffix, and the fetch_... key goes in apiKey. Then pick a model via /models → Anthropic. A step-by-step guide is in connecting opencode to an API gateway. The current model list with prices is available from GET https://api.zukko.pro/v1/models.

What an hour of agent work costs

Token usage depends on the project, so take a hypothetical hour: the agent reads 3M input tokens and writes 150K output tokens. This is an assumption for illustration, not a measurement. Prices are from the Zukko catalog, per 1M tokens.

ModelInputOutputTotal for the hour
claude-haiku-5-53 × $0.05 = $0.150.15 × $0.2 = $0.03$0.15 + $0.03 = $0.18
deepseek-v4.1-flash3 × $0.03 = $0.090.15 × $0.1 = $0.015$0.09 + $0.015 = $0.105
glm-5.3-flash3 × $0.04 = $0.120.15 × $0.04 = $0.006$0.12 + $0.006 = $0.126
claude-sonnet-53 × $0.4 = $1.200.15 × $2 = $0.30$1.20 + $0.30 = $1.50
claude-opus-5-53 × $0.8 = $2.400.15 × $4 = $0.60$2.40 + $0.60 = $3.00

The Zukko docs for opencode cover Claude models. The gateway accepts other catalog models in the same format (that is how they work in Claude Code); add their ids to the same models block, but test a given model in opencode on a small task first.

Input tokens make up most of the bill because the agent keeps rereading files and history. Ways to cut that are in the token-saving checklist. For GLM 5.3 Flash specifically, see GLM API free.

Payment

Zukko keeps a balance in dollars and charges only for tokens used, with no subscription fee. Top-ups start at $1: via SBP (Russia's Faster Payments System) in rubles at the market rate (around 8.5% added on the payment page), cryptocurrency with no surcharge, or an international MasterCard ($2 to $594 per payment, around 13% added). Russian bank cards are temporarily not accepted; ruble payments go through SBP.

There is no free tier and no sign-up bonus. A one-time $0.50 is given for subscribing to the @zukkopro Telegram channel. Each key can have daily and monthly spending limits in dollars, so the agent cannot spend more than planned.

FAQ

Is opencode completely free?

The agent is, under the MIT license. You pay only for models. The built-in provider also has free models, but they are free for a limited time.

What is Big Pickle in opencode?

A stealth model with an undisclosed name, offered free through the built-in provider. While it is free, data may be used to improve it. Its config ID is opencode/big-pickle.

How long will free access last?

Unknown. The docs only say "for a limited time", with no dates or request limits.

Can I use opencode's free models from another client?

No. Free chat models work only inside OpenCode. Other clients need paid models, a Go subscription, or your own provider.

Is it safe to send work code to the free models?

It depends on the model. GLM-5.3-Flash Free and DeepSeek V4.1 Flash Free state no storage and no training. Proprietary code is better kept out of Big Pickle, Flint 40B Free, and the Nemotron models.

zukko

Pay 25×
less

for Claude Code & ChatGPT

One fetch_* key and base URL api.zukko.pro — for Cursor, Claude Code, Codex, and any SDK.

Get API keyDocumentation

api.zukko.pro · fetch_*

Opencode free models: Big Pickle, limits and data terms