Crouton Alpha and Pulse Alpha: Status, Privacy, and Migration

OpenCode lists Crouton Alpha Free as deprecated from November 9, 2026, and Pulse Alpha Free from November 6. Review their catalog limits without confusing metadata with availability, distinguish no-training claims from retention permissions, and replace preview IDs in Claude Code and OpenCode. Includes gateway authentication, persistent defaults, a file-read check, and guidance for quota failures and paid alternatives.

Contents
Crouton Alpha and Pulse Alpha: Status, Privacy, and Migration

Replace Crouton Alpha Free and Pulse Alpha Free in OpenCode rather than starting a new workflow on those previews. As of 2026-11-10, OpenCode lists Crouton as deprecated from November 9 and Pulse from November 6. Select an available replacement and verify a real file-read task before resuming coding.

Are Crouton Alpha and Pulse Alpha still available?

OpenCode’s deprecation list is the deciding source for its free routes. Its official model documentation records:

OpenCode previewModel IDDeprecation date
Crouton Alpha Freecrouton-alpha-freeNovember 9, 2026
Pulse Alpha Freepulse-alpha-freeNovember 6, 2026

These are OpenCode’s dates. They do not establish the lifecycle of stealth/crouton-alpha on OpenRouter or another provider’s endpoint.

For an OpenRouter deployment, check its model-specific status before relying on an old setup guide. Do not treat a catalog entry, a working login, or an unused balance as proof that a preview still accepts inference requests.

“Deprecated” also does not establish the exact error every client will return. Check the provider notice before spending time on authentication changes or waiting for a daily reset.

What do the 1M-context specifications establish?

The catalog limits describe the previews, not continuing availability. The operator-supplied records in OpenCode’s model database contain these values:

OpenCode modelContext tokensOutput-token limitCatalog input types
Crouton Alpha Free1,048,576270,000Text, image, video, PDF
Pulse Alpha Free1,048,576131,072Text, image

Both records now have status = "deprecated". These are OpenCode-specific limits; do not transfer them to OpenRouter merely because the preview name matches.

A context limit is not an output allowance, a daily message quota, or a coding-quality score. A client may also impose a smaller working context than the provider’s catalog allows.

Do not choose a replacement by guessing which laboratory produced a stealth model. Select a documented model ID, then compare its behavior on your actual task.

Did the previews retain data or use it for training?

Crouton’s launch announcement and Pulse’s preview policy made different promises. OpenRouter’s November 3 announcement advertised no training on submitted data. That statement alone does not promise zero retention.

OpenCode’s original Pulse documentation allowed collected data to be retained and used to improve the model. Treat that as permission for model improvement, not as a no-training guarantee.

Pulse’s documented preview exclusions covered the European Union, Iceland, Liechtenstein, Norway, Cuba, Iran, North Korea, Russia, Ukraine, and Venezuela. Those historical restrictions do not establish the eligibility rules of a replacement.

Check retention, training or improvement, and regional eligibility separately before reconnecting a private repository. Data-location requirements are another check: routing inside the EU and allowing EU users are different conditions.

Use public or redacted files for the initial trial. Exclude credentials, customer records, and confidential logs, including material a file-reading tool could discover automatically.

How do you replace a preview in Claude Code?

Keep the gateway configuration explicit and replace the model selector before permitting edits. This example uses the paid anthropic/claude-sonnet-5.5 entry in OpenCode’s OpenRouter catalog.

It is a named migration candidate, not a claim about either preview’s identity or a benchmark winner. You need Claude Code installed, your own OpenRouter API key, and credits for paid requests. These requests use OpenRouter billing rather than a Claude subscription; see the Claude Code installation and billing guide for that distinction.

Start in a repository containing a non-sensitive README.md. If Claude Code has a cached Anthropic login, run /logout in that client and quit before switching authentication.

The following Bash example follows the OpenRouter integration guide. Enter your key at the hidden prompt:

read -rs OPENROUTER_API_KEY && export OPENROUTER_API_KEY

(
  export ANTHROPIC_BASE_URL="https://openrouter.ai/api"
  export ANTHROPIC_API_KEY="$OPENROUTER_API_KEY"
  export ANTHROPIC_AUTH_TOKEN=""
  export ANTHROPIC_DEFAULT_SONNET_MODEL="anthropic/claude-sonnet-5.5"
  export CLAUDE_CODE_SUBAGENT_MODEL="anthropic/claude-sonnet-5.5"
  claude --model "anthropic/claude-sonnet-5.5"
)

The parentheses scope these settings to the launched process. They do not rewrite your shell profile or project configuration.

The base URL ends in /api, not /api/v1. This configuration puts the OpenRouter key in ANTHROPIC_API_KEY and explicitly empties ANTHROPIC_AUTH_TOKEN. Do not combine it with a conflicting authentication recipe.

Run /status inside Claude Code. Check for ANTHROPIC_API_KEY authentication and the base URL https://openrouter.ai/api. Open /model to inspect the selected model.

Then submit this task:

Use a file-read tool to read README.md. Return its first heading exactly, then summarize the project's purpose in two sentences. Do not edit files or run commands.

Look for an actual file-read result and the correct heading. Check that no files changed. Also inspect the request’s model in the OpenRouter Activity dashboard; the model’s self-description is not routing evidence.

OpenRouter recommends prioritizing the Anthropic first-party provider for Claude Code compatibility. Follow that recommendation before treating another upstream route as interchangeable.

Before a larger task, inspect existing role overrides for the old preview ID. Check ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_HAIKU_MODEL, and ANTHROPIC_DEFAULT_FABLE_MODEL, as well as Sonnet and subagent settings.

The Claude Code model reference distinguishes startup selection from role overrides. Change only stale entries. A successful main-session request does not verify every separately configured helper.

How do you replace Pulse or Crouton in OpenCode?

Update both the active session and the persistent default. The example below reuses the OpenRouter account from the Claude Code setup.

Start opencode in the test repository. Run /connect, select OpenRouter, and enter your OpenRouter key. Run /models and choose openrouter/anthropic/claude-sonnet-5.5.

OpenRouter’s OpenCode integration guide documents this connection flow. The first openrouter/ identifies OpenCode’s provider; the remaining anthropic/claude-sonnet-5.5 is the gateway’s model ID.

After connecting, run this Bash check:

opencode run --model openrouter/anthropic/claude-sonnet-5.5 \
  "Use a file-read tool to read README.md. Return its first heading exactly, then summarize the project's purpose in two sentences. Do not edit files or run commands."

Review the tool result and reply in the terminal. Verify the heading, confirm no edits, and check the request’s selected model in OpenRouter Activity.

To set the default for new work, use OpenCode’s documented top-level model setting. Merge this structure into the project’s opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "openrouter/anthropic/claude-sonnet-5.5"
}

Keep unrelated settings intact. OpenCode’s configuration guide confirms the $schema value and documents project and user-wide locations.

An existing session’s selected model takes precedence over the configured default. Use /models in that session too. Selecting a session model does not rewrite the configuration, as the current model-selection documentation explains.

If the replacement is missing, check that its provider and model are enabled in this project. Restart after connecting a provider when necessary. Do not guess another model ID or copy one from a different provider.

Do not move Pulse’s old endpoint into Claude Code as a workaround. OpenCode’s model documentation restricts free-tier chat models to OpenCode. An API-compatible endpoint does not remove that client requirement.

Is a failed request a quota or configuration problem?

Check lifecycle status first, then authentication and allowances. A daily reset and a model deprecation are different events.

For authentication failures, inspect the key source and base URL. In Claude Code, recheck /status and cached login state. In OpenCode, reconnect the intended provider and verify the full provider/model selector.

For a free-message limit, use the allowance and reset information shown for your account. Do not generalize a reset time from another user’s screenshot. Keep daily allowances separate from request-rate limits and provider capacity.

For timeouts or empty responses, save the task state before retrying. Do not start several parallel attempts against an uncertain route. After a bounded retry, use an approved replacement rather than leaving an unattended loop running.

Launch-day availability percentages cannot establish today’s service status. For a replacement trial, record completed tasks, failures, retries, and billed usage. A text response alone does not prove that file access, tool calls, and the rest of the coding workflow work.

When should you choose a paid replacement?

Choose a named paid model when preview interruptions block work or the preview’s data terms fail your requirements. Payment alone does not guarantee availability, regional access, or zero retention.

The Sonnet configuration provides one concrete candidate. To evaluate another route, use the MiniMax M3.1 Flash setup guide and repeat the same acceptance task.

Check current token prices, cache rules, and mandatory platform fees before adding credit. Compare charges for the same accepted result, including retries. Track review time separately rather than declaring a winner from token prices alone.

Before adopting another free preview, save unfinished work locally and keep an approved replacement configured. Record the exact provider/model ID and data terms. Return to a preview only when its current eligibility and behavior pass the same checks.

Ready to optimize your LLM workflow?

Join thousands of developers building faster, smarter, and more cost-effective AI applications with BetterToken.

Get Started for Free