---
title: "OpenCode Free Model Errors: v1 Shutdown, v2 Upgrade, and Reset Delays"
description: "Since OpenCode shifted to v2, free-tier errors have piled up in #59387. Here is the context, the exact error messages, and each of the four root causes explained one by one."
url: https://codingpass.ai/blog/opencode-free-model-errors-v2-upgrade
language: en
updated: 2026-11-20
---
# OpenCode Free Model Errors: v1 Shutdown and v2 Upgrade

Reports of OpenCode free models stopping mid-session surged in #59387 after the v2 rollout. Many users assume they hit the daily cap and reach for a Go subscription, yet at least four separate causes trigger similar popups.

[CodingPass team](https://codingpass.ai/about) Published Nov 20, 2026 Updated Nov 20, 2026 6 min read

In late October, OpenCode rolled out 2.x and began shutting off free-model access on older clients. Reports of broken free models quickly piled up on GitHub, prompting maintainers to designate [anomalyco/opencode#59387](https://github.com/anomalyco/opencode/issues/59387) as the tracking thread and close new tickets as duplicates.

The typical sequence looks straightforward at first: a session stops mid-task with a red banner saying free usage is exceeded. Users follow a quick guide to update, or wait for the stated reset timestamp, only to see the reset date slip forward by another day. Although the symptom sounds like "out of free quota," four distinct issues sit behind it.

Check the exact wording before changing settings or subscribing to Go. Outdated 1.x clients, legacy 1.x sessions, stuck server counters, and non-OpenCode clients each require a different fix.

## What the errors look like

The popups look nearly identical, so the exact text and buttons carry all the diagnostic clues:

```
1. This version of OpenCode Desktop no longer supports free models.
2. Free usage exceeded, subscribe to Go
3. You've used today's free messages. They reset at Oct 22, 2026, 8:00 p.m.
4. FreeUsageLimitError (HTTP 429)
```

When you see the third message with `Upgrade to OpenCode Go` and `Switch model` buttons underneath, your client has already passed the version check and the block is on the quota side. The other messages have nothing to do with running out of daily messages.

## The four possible causes

### 1. You are still on 1.x, where free models are shut down

Seeing `no longer supports free models` or `Free usage exceeded, subscribe to Go` almost always means your client is still on 1.x. Maintainers confirmed on October 27 that v1 no longer supports free models.

Many users stay stuck on 1.x while assuming they are up to date for two reasons. First, as noted in `#59482`, `The v1 Desktop's built-in updater keeps you on 1.x, so updating from inside the app won't get you there.` Second, OpenCode has two npm packages: the legacy `opencode-ai` package stops at `1.19.5`, whereas the v2 release lives under `@opencode/cli` (`2.2.2`+ on `latest`). Running `npm i -g opencode-ai` leaves you on 1.x indefinitely.

Because Desktop and CLI are separate binaries, verify `2.x` in the interface you actually use: `opencode --version` in your terminal, `Settings → About` in the desktop app, or the `.exe` file version in your Windows install folder. To install v2 manually, pick the path for your platform:

| Method | Platform | Command or steps |
| --- | --- | --- |
| curl | macOS / Linux | `curl -fsSL https://opencode.ai/v2/install \| bash` |
| npm | All platforms | `npm install -g @opencode/cli` |
| Homebrew | macOS | `brew install anomalyco/tap/opencode-v2` |
| Standalone binary | Windows | Download the official x64 zip, extract it, and add the folder to `PATH` |

The official docs note that Windows package managers are not supported, so Windows users should use npm or the standalone zip. After installing, if `/models` shows no free models, run `/connect` to sign in to your OpenCode Console account—the built-in `opencode` provider does not appear while logged out.

### 2. You upgraded to v2, but are testing inside a v1 session

After installing v2, many users reopen an existing thread to continue their task and still hit `no longer supports free models`, leading them to assume the upgrade failed.

Issue `#60352` documents this exact behavior: even when both CLI and desktop are on `2.2.2` and local database migration succeeds, the backend still rejects **sessions created back in 1.x** as outdated clients. Opening a **brand-new session** with the same account and free model usually works immediately. Always verify a v2 upgrade in a new session first.

### 3. Daily quota is exhausted, or the server counter is stuck

When `You've used today's free messages. They reset at <time> · <date>` appears, watch whether the reset time stays fixed. If it points to the same hour every day, you genuinely hit the daily free cap and just need to wait for the reset.

If the countdown hits zero and immediately jumps forward by another day (for example, from `Oct 22` to `Oct 23`), the server counter is stuck. Per `#60328`, quotas and reset windows are controlled by the hosted Console backend per account and network, so local tweaks cannot fix them. Avoid reinstalling or deleting config folders and databases—that rarely resolves the counter and wipes your local session history. Two unofficial workarounds reported by users are signing out of Console inside OpenCode and signing back in (sometimes more than once), or switching networks such as a mobile hotspot.

### 4. You are calling free models from outside OpenCode

If you get `FreeUsageLimitError` or `HTTP 429` from another editor or a custom script, version and network have nothing to do with it. OpenCode's documentation explicitly states `Free-tier chat models require OpenCode`. Free-tier keys only work inside the OpenCode client; external callers receive HTTP 429 regardless of setup. Switch back to OpenCode or use a paid model instead.

## Two edge cases that look worse than they are

- **402 logs from background title requests**: Seeing 402 rejections for low-cost models like `GPT-6 Luna` in local logs does not mean OpenCode is charging you silently. Console docs explain that OpenCode uses low-cost models (Haiku, Nano, Flash) on a separate background channel to generate session titles, independent of your free chat quota.
- **`Illegal instruction` on older CPUs**: CPUs over a decade old that top out at SSE4.1 (missing SSE4.2 and POPCNT) crash on v2 with `Illegal instruction`, while v1 no longer supports free models. That is a hardware instruction-set limit rather than a config bug.

If you are still blocked after verifying `2.x`, testing a new session, and trying a re-login or network switch, report in `#59387` with three items: your exact version, OS, and full error text.

## When OpenCode Go is worth buying

The OpenCode Go page notes: `If you reach the usage limit, you can continue using the free models.` Go buys paid quota on a separate pool—it does not unlock free models.

Subscribing to Go will not bypass a 1.x client, a legacy 1.x session, or a third-party 429. It only makes sense when you are on a fresh v2 session, the reset time stays fixed, and you routinely exhaust the daily free cap mid-day (Go is `$10/mo`, Go Plus `$40/mo`, and Go Teams `$20/user/mo`; OpenCode does not publish an exact daily message limit for free models).

**Pool and regional notes**: The current limited-time free pool includes Big Pickle, GLM-5.3-Flash Free, DeepSeek V4.1 Flash Free, Beam Free, Flint 40B Free, MiMo-V2.6-Flash Free, Nemotron 3 Ultra Free, Nemotron 3.5 Lightning Free, Nemotron 3.5 Super VL Free, and Jev 1.13 Free. **Beam Free is not available in China (including Hong Kong and Macau)**. To hide free models altogether, set `OPENCODE_DISABLE_FREE_MODELS=1`.

Written by CodingPass team

We run CodingPass and test each guide against the latest stable release of the tool before we publish it. Spotted something out of date? Write to [support@codingpass.ai](mailto:support@codingpass.ai).

## Related reading.

Guide

### [OpenCode setup guide](https://codingpass.ai/tools/opencode)

Provider config in opencode.jsonc, model IDs per plan, and how to check common errors.

Rules

### [How allowances and caps work](https://codingpass.ai/usage-limits)

Allowances and caps for every plan and model, plus the 429 response.

Models

### [All models](https://codingpass.ai/models)

Model IDs, required tiers, and monthly allowances.
