Skip to main content
When you submit a hosted agent task with one of the older model names below, Hyperbrowser accepts that name as an alias and replaces it before starting the task. For example, requesting gpt-4.1-nano with HyperAgent runs gpt-6-luna. Mappings depend on the agent endpoint. The same name can select different replacements: claude-sonnet-4-5 maps to claude-sonnet-4-6 in HyperAgent and Browser Use, but to claude-sonnet-5-5 in Claude Computer Use. Use the replacement name directly in new integrations.
These tables describe aliases applied by the hosted Hyperbrowser API. They do not configure the standalone HyperAgent package, locally run agents, or model aliases managed by a provider.

​
How mappings affect your requests

  • You can continue submitting the listed aliases, but they select the replacement model rather than the original model.
  • Mappings also apply when you use your own API keys. Your provider account must have access to the replacement model.
  • If you omit a model field, the endpoint applies its default and then resolves any alias. See each agent’s parameter reference for its defaults.
  • Model changes can affect output, latency, step counts, and cost. Test representative tasks when migrating, and review pricing. With your own keys, the provider bills you for the model used.
  • A model name absent from these tables has no alias listed here; it still needs to be supported by the agent and provider. These tables are not a complete list of supported models.

​
HyperAgent

Applies to llm on POST /api/task/hyper-agent. See the HyperAgent guide. The OpenAI mappings apply to both HyperAgent 0.8.0 and 1.1.0. Claude and Gemini require 1.1.0; aliases do not bypass that restriction. The default model remains gpt-4o when you omit llm. For the nano migration, you can keep sending llm: "gpt-4.1-nano" or explicitly select llm: "gpt-6-luna". The hosted runtime handles Luna’s request settings for both HyperAgent versions. You do not need to add a reasoning parameter to your task request.

​
Browser Use

Applies separately to llm, plannerLlm, and pageExtractionLlm on POST /api/task/browser-use. See the Browser Use guide. These mappings apply to versions 0.1.40, 0.7.10, and latest. The API defaults to version 0.7.10. Each omitted model field defaults to gemini-2.0-flash, which resolves to gemini-3-flash-preview. For platform-billed requests, a Browser Use model override configured for your team takes precedence. Requests using your own API keys retain the mappings above.

​
Claude Computer Use

Applies to llm on POST /api/task/claude-computer-use. See the Claude Computer Use guide.

​
OpenAI CUA

Applies to llm on POST /api/task/cua. See the OpenAI CUA guide. If you omit llm, the default is gpt-5.6-terra.

​
Gemini Computer Use

Applies to llm on POST /api/task/gemini-computer-use. See the Gemini Computer Use guide.

​
Grok Computer Use

Applies to llm on POST /api/task/grok-computer-use. See the Grok Computer Use guide.

​
Agents without API model mappings

Meta Computer Use and Jev Computer Use currently have no automatic model replacements in their task endpoints. Refer to their guides for supported model names.