Switching Claude Code Between Backends on One Aggregator Key: What Changes in the Session

What stays the same when you switch backends

Your Claude Code session is primarily a client-side construct. The conversation history, tool definitions, and local file context live in your environment, not on the backend. When you point Claude Code at a different model through the same aggregator key, the following remain unchanged:

• Your API key and endpoint – The aggregator key authenticates all requests; you only change the model identifier.
• Conversation history – Messages already exchanged are still part of the context sent with each new request.
• Local tooling – File system access, shell commands, and other tools run in your environment and are unaffected by the backend swap.
• Billing relationship – You continue to pay the aggregator’s uniform multiplier (official price × 1.3) regardless of which model you call.

What changes in the session

Switching backends alters the model that generates responses. This has several practical effects:

• Model-specific behavior – Different models have different strengths, token limits, and response styles. Code generation, reasoning depth, and instruction following can vary.
• Context window size – Each model has its own maximum context length. A model with a smaller window may truncate earlier messages or require you to start a fresh session.
• Tokenization and cost – Token counts differ between models, so the same conversation may consume a different number of tokens. Your cost is still calculated as official price × 1.3, but the underlying token count affects the final amount.
• Latency and throughput – Response times and rate limits depend on the model and the aggregator’s routing.

How to switch backends

In Claude Code, the backend is typically selected via a configuration setting or environment variable that specifies the model name. With an aggregator key, you use the aggregator’s model identifier (for example, claude-3-5-sonnet or gpt-4o) in the same way you would use a provider-specific name.

1. Locate your Claude Code configuration (often a .claude/config.json file or environment variables like ANTHROPIC_MODEL).
2. Replace the model identifier with the desired model from the aggregator’s supported list.
3. Restart the Claude Code session or reload the configuration.

The aggregator key remains the same; only the model string changes.

Managing context and continuity

When switching to a model with a smaller context window, you may need to trim the conversation history. Claude Code does not automatically summarize or forget messages; it sends the full history up to the model’s limit. To avoid errors:

• Start a new session if the history exceeds the new model’s window.
• Use Claude Code’s built-in commands to clear or compact history if available.
• Keep critical context in local files or notes that you can re-inject.

Billing and contributor credits

All calls through the aggregator are billed at official price × 1.3, regardless of the backend. If you are a key contributor (for example, you provided the API key or refer users), you receive credits at official price × 1.1 (or × 1.2 for premium models) in USDC. Switching backends does not change your contributor rate.

Practical tips

• Test a new backend with a small, non-critical task before committing to a long session.
• Monitor token usage when switching, as different models may tokenize the same text differently.
• Keep your aggregator key secure; it works across all supported models.
• Refer to the aggregator’s documentation for the exact model identifiers and any model-specific parameters.