GPT-6.1 Sol on Hermes Agent and OpenClaw

What GPT-6.1 Sol is

OpenAI released gpt-6.1-sol on September 29, one week after GPT-6 Sol. OpenAI describes it as "near-Astra performance for complex work at a lower cost" and recommends it for complex coding, computer use, and professional work, such as building a website from a product brief or turning financial results into a board presentation.

At the DevDay launch, OpenAI said it matches GPT-6 Astra on the DeepSWE v1.1 software-engineering test, beats Claude Opus 5.5 on AutomationBench and Terminal-Bench Science, and lands 2.1 points behind Astra on OSWorld 2.0 computer use at about a seventh of the cost. These are vendor results. OpenAI's own guidance is to compare it with Astra on your tasks before choosing.

It also supports multi-agent delegation in beta on the Responses API, where the model hands parts of a request to subagents.

Use GPT-6.1 Sol with your ChatGPT subscription

If you already pay for ChatGPT Plus, Pro, or a business plan, this is the cheapest way to give an always-on agent GPT-6.1 Sol. Turns count against your ChatGPT plan's own limits instead of an API bill.

1. Deploy a Hermes Agent on OpenClaw Launch, or open an existing one from your Dashboard.
2. Click Terminal on the instance and run hermes auth add openai-codex --type oauth.
3. Open https://auth.openai.com/codex/device in your own browser, enter the code the terminal shows, and sign in with your ChatGPT account.
4. Open the instance card's model menu and pick GPT-6.1 Sol under the OpenAI row badged Subscription, or switch from any chat with your bot:

   /model gpt-6.1-sol --provider openai-codex --global

Hermes and the instance card read the model list live from your ChatGPT account, so GPT-6.1 Sol appears as soon as your plan serves it. If you also use the Codex CLI directly, update it to the latest release first. In our test, Codex CLI 0.155.1 did not list GPT-6.1 Sol and rejected it with "not supported when using Codex with a ChatGPT account", on an account whose catalogue did include it for newer clients. This route is Hermes-only for now; OpenClaw offers only the subscription models its own release already knows. The full walkthrough is in the Hermes ChatGPT subscription guide.

Deploy Hermes Agent See plans

Pick GPT-6.1 Sol with an API key on a managed instance

With an API key or OpenRouter credits, the steps are the same on Hermes Agent and OpenClaw, and you do not need a new instance.

1. Open the Dashboard and find your running instance.
2. Click the current model in the Text or Model row of the instance card.
3. Choose the provider row that should pay:
   • OpenAI with the Your key badge, if you have saved an OpenAI API key on the API Keys page. Tokens are billed to your OpenAI account.
   • OpenRouter, to use your plan's included credits or your own OpenRouter key. The model appears as openai/gpt-6.1-sol.
4. Search for 6.1 sol and select it. The OpenAI list picks up new OpenAI models automatically, so GPT-6.1 Sol is there even though the featured shortlist still ends at GPT-6 Sol.
5. Wait for the instance to return to ready, then start a fresh conversation.

Full-catalogue access applies to eligible paid instances; trial accounts use a restricted model list. On OpenClaw, a pick from the OpenAI row is registered with the Responses API automatically, which GPT-6.1 Sol needs for tool calls. Hermes talks to api.openai.com over Responses already.

Pricing compared with GPT-6 Sol and Astra

All three have a 1,050,000-token context window and 128,000 maximum output tokens. GPT-6.1 Sol accepts up to 922,000 input tokens. Prompts above 272,000 input tokens are billed at 2x input and cache rates and 1.5x output for the whole request. Batch and Flex cost half of Standard, and Fast mode costs 2x. Prices are from the GPT-6.1 Sol model page, checked September 29, 2026.

The cached-input price matters most for agents. A long-running agent resends the same system prompt, tools, and history on every turn. At $0.10 instead of $0.20, those repeated tokens cost half as much as on GPT-6 Sol. OpenRouter lists openai/gpt-6.1-sol at the same $2, $10, and $0.10.

Two changes that can break an existing setup

Reasoning cannot be turned off

GPT-6.1 Sol accepts low, medium (the default), high, xhigh, and max. GPT-6 Sol also accepted none, which turned reasoning off; GPT-6.1 Sol rejects it, and it rejects minimal too. OpenRouter's catalogue still lists none for this model, but OpenAI rejects it there as well. Current Hermes omits an unsupported none, so the request succeeds but the model reasons at its default medium rather than the fast, no-thinking turn you asked for; older Hermes builds may send it and fail. Either way, if you had set agent.reasoning_effort to "none" for speed, change it to "low", the lightest setting this model has. Managed OpenClaw Launch instances leave reasoning effort at the default and are not affected unless you changed it.

Tool calls need the Responses API

Chat Completions works only for requests without tools. An agent is all tool calls, so a custom integration that sends tools to /v1/chat/completions fails on GPT-6.1 Sol even though the same call worked on GPT-6 Sol with reasoning none. Hermes and OpenClaw handle this on their direct OpenAI routes, as described below. It matters mainly if you point either one at an OpenAI-compatible relay that only speaks Chat Completions.

Self-hosted Hermes Agent

Put the key in ~/.hermes/.env, then set the provider and the bare model ID in ~/.hermes/config.yaml:

# ~/.hermes/.env
OPENAI_API_KEY=sk-...

# ~/.hermes/config.yaml
model:
  provider: "openai-api"
  default: "gpt-6.1-sol"

agent:
  reasoning_effort: "medium"   # low is the lightest; "none" is not accepted

Hermes' openai-api provider uses the Responses API, so tools work. Through OpenRouter instead, use OpenRouter's slug:

model:
  provider: "openrouter"
  default: "openai/gpt-6.1-sol"

You can also run hermes model and pick it from the list, since current upstream Hermes lists gpt-6.1-sol for both providers. On an older build that does not list it, type the ID in the custom model field. A model change applies to new sessions. To restart the gateway, run hermes gateway restart. Field names follow the upstream configuration example.

Self-hosted OpenClaw

No OpenClaw release so far, up to 2026.9.6, includes gpt-6.1-sol in its built-in OpenAI catalogue. The model is on OpenClaw's unreleased main branch, so a coming release should pick it up. Until then, set up an OpenAI API-key profile and register the model yourself:

openclaw onboard --auth-choice openai-api-key
openclaw models list --provider openai

Add this entry to your existing models.providers.openai.models array. Setting api to openai-responses makes tool calls go over the Responses API, which GPT-6.1 Sol requires. reasoning: true tells OpenClaw it is a reasoning model, so it defaults to medium and accepts /think low:

{
  "models": {
    "providers": {
      "openai": {
        "models": [
          {
            "id": "gpt-6.1-sol",
            "name": "GPT-6.1 Sol",
            "api": "openai-responses",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 1050000,
            "maxTokens": 128000
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": { "primary": "openai/gpt-6.1-sol" }
    }
  }
}

Merge this into your configuration rather than replacing the file. Avoid /think off on this model: a release that does not know GPT-6.1 Sol cannot know that it rejects a reasoning disable. Use /think low for the lightest setting. Through OpenRouter, the reference is openrouter/openai/gpt-6.1-sol after openclaw onboard --auth-choice openrouter-api-key. See the upstream OpenAI models guide for your installed version.

GPT-6.1 Sol, GPT-6 Sol, or Astra?

For a coding or research agent that plans and uses tools across many steps, start with GPT-6.1 Sol. It costs the same per token as GPT-6 Sol and is cheaper on cached context. Keep GPT-6 Sol or Luna for setups that depend on reasoning none, and Luna for high-volume, narrow tasks. Use Astra when a wrong answer costs more than the 5x price difference. Run the same real tasks on each and compare cost per finished task, not per token.

GPT-6.1 Sol FAQ

Related model guides

• GPT-6 Sol and Luna on Hermes Agent
• GPT-6 Astra setup
• Hermes Agent with a ChatGPT subscription
• Hermes Agent with your own API key