Claude Max Plan banned - alternative Sisyphus model options?

[callum-spencer]

BELOW IS MODIFIED BY @code-yeongyu :

I HAVE NOT CREATED A SINGLE IMPLEMENTATION RELATED TO ANTHROPIC OAUTH.

I HAVE 3X MAX20 - AND I HAD NO PROBLEMS.

THEREFORE, PLEASE STOP ASKING ME IF USING OH MY OPENCODE WILL GET YOUR CLAUDE ACCOUNT BANNED.
I AM STRESSING THIS ONCE AGAIN: I DO NOT HAVE A SINGLE OAUTH-RELATED IMPLEMENTATION.

---

ORIGINAL:

As the title suggests, after heavy use with the Sisyphus agent using opus 4.5 (Just on usual coding tasks), my claude max account has been banned & I've received a refund, what would you suggest I use as the model for Sisyphus instead? Tried GPT 5.2 (extra high), just want to make sure this is the recommended option..

Appreciate any help or suggestions you can provide, as I assume this will becoming a reoccurring problem and people will likely need additional options for different models to use with Sisyphus.

[code-yeongyu]

5.2 is a good take.

but just curious - what made you to get banned? how did you use it ? i feel so sorry about it that btw

[kevincojean]

I've had good results using sonnet 4.5 with thinking at high

[oritwoen]

You can subscribe to GitHub Copilot and use claude-opus-4.5 just like claude max, but change the provider in the agent from anthropic to github-copilot. I've been using this for two days.

[callum-spencer]

5.2 is a good take.

but just curious - what made you to get banned? how did you use it ? i feel so sorry about it that btw

I wasn't actually doing anything out of "ordinary" usage, literally just debugging lint failures after upgrading NextJS version lol for my web app..

I found a decent workaround, I have the Google AI Ultra plan, so I'm using opencode-antigravity-auth@1.1.2 and using opus 4.5 for Sisyphus via that.. beats me paying for multiple providers, so can just stick with ChatGPT Plus + Google AI Ultra now

[callum-spencer]

You can subscribe to GitHub Copilot and use claude-opus-4.5 just like claude max, but change the provider in the agent from anthropic to github-copilot. I've been using this for two days.

Yeah I pretty much stumbled on a similar workaround, I just use my antigravity auth to access opus 4.5 instead

[seanGSISG]

how long had you been a max subscriber? do you use vpn's at all?

how did you have oh-my-opencode agents configured? were you just using opus 4.5 for the main agent?

[code-yeongyu]

Glm 4.7 definitely

[waynebrantley]

@callum-spencer it is a terms of service violation to use anything but claude code with their subscription.
Try google ai pro/ultra - it gives you all you need.

[ElderOrb]

in a case if opensource-only models, which ones from the list are good and for which role: Deepseek 3.2 ? Minimax 2.1, Glm 4.7, Xiaomi MIMO ?

[Straffern]

in a case if opensource-only models, which ones from the list are good and for which role: Deepseek 3.2 ? Minimax 2.1, Glm 4.7, Xiaomi MIMO ?

I have been testing with GLM-4.7 and Minimax M2.1
Both seems incapable of delegating work to other agents and also did not use any of the tools such as lsp_*.

[shawnmarck]

in a case if opensource-only models, which ones from the list are good and for which role: Deepseek 3.2 ? Minimax 2.1, Glm 4.7, Xiaomi MIMO ?

I have been testing with GLM-4.7 and Minimax M2.1 Both seems incapable of delegating work to other agents and also did not use any of the tools such as lsp_*.

were you starting your session with these models or switching to them? I've noticed some inconsistencies depending on that.

[nghyane]

You can integrate llm-mux to multiplex multiple Antigravity accounts and effectively increase your quota limits.

Below is a full example showing how to route Anthropic and Gemini traffic through the local llm-mux gateway.

Provider Configuration

{
  "provider": {
    "llm-mux-claude": {
      "npm": "@ai-sdk/anthropic",
      "name": "LLM-Mux Claude",
      "options": {
        "baseURL": "http://localhost:8317/v1",
        "apiKey": "sk-dummy-key"
      },
      "models": {
        "claude-opus-4-5-thinking": {
          "id": "claude-opus-4-5-thinking",
          "name": "Claude 4.5 Opus Thinking",
          "tool_call": true,
          "temperature": true,
          "reasoning": true,
          "limit": {
            "context": 200000,
            "output": 64000
          },
          "options": {
            "maxOutputTokens": 64000,
            "thinking": {
              "type": "enabled",
              "budgetTokens": 32000
            }
          }
        }
      }
    },

    "llm-mux-gemini": {
      "npm": "@ai-sdk/google",
      "name": "LLM-Mux Gemini",
      "options": {
        "baseURL": "http://localhost:8317/v1beta",
        "apiKey": "sk-dummy-key"
      },
      "models": {
        "gemini-3-pro-preview": {
          "id": "gemini-3-pro-preview",
          "name": "Gemini 3 Pro",
          "tool_call": true,
          "temperature": true,
          "reasoning": true
        },
        "gemini-3-flash": {
          "id": "gemini-3-flash",
          "name": "Gemini 3 Flash",
          "tool_call": true,
          "temperature": true,
          "reasoning": true
        }
      }
    }
  }
}

Agent Mapping

{
  "agents": {
    "Sisyphus": {
      "model": "llm-mux-claude/claude-opus-4-5-thinking"
    },
    "librarian": {
      "model": "llm-mux-gemini/gemini-3-flash"
    },
    "explore": {
      "model": "llm-mux-gemini/gemini-3-flash"
    },
    "oracle": {
      "model": "llm-mux-claude/claude-opus-4-5-thinking"
    },
    "frontend-ui-ux-engineer": {
      "model": "llm-mux-gemini/gemini-3-pro-preview"
    },
    "document-writer": {
      "model": "llm-mux-gemini/gemini-3-flash"
    },
    "multimodal-looker": {
      "model": "llm-mux-gemini/gemini-3-flash"
    }
  }
}

By routing requests through llm-mux, you can:

• Seamlessly rotate across multiple Antigravity accounts
• Avoid per-account quota bottlenecks
• Maintain a unified API interface
• Keep existing agent definitions unchanged while benefiting from distributed load

[faction23]

Also got banned friday with this setup (claude max, opencode, oh-my), just using a single agent, no vpn. The work was just overhauling a storybook to enhance stories and upgrade to sb 10. I only found out afterwards that it was a tos issue to use the plan with opencode apparently. Def important to make people a bit more aware on docs, though in the end, my bad.

That said i have used claude max plans with opencode before oh my for quite a few months. And this setup had worked well for a full week on a daily basis before the completely random appearing ban based on my activity. Many people are getting banned over the last year for what they claim are false positives, though yes, many of those seem to have been related to vpn's. So its not 100% to make the correlation that opencode plus oh my is causing any bans.

The fact anthropic do lifelong bans with a broken appeals process seems like a bad move on their part, some kind of suspension and appeal with one grace and clear explanation at least would seem better for their business hahaha, but, i guess its their choice. I dont really see how using the max sub through opencode is in any way harmful to them, usage is usage.

Also using antigravity approach now.

[waynebrantley]

@faction23 I agree with you. Same situation. Appealed weeks ago, no word.