Summary
This should be treated as more than a Windows crash bug.
MiniMax Code 3.0 may look strong on paper: AI coding agent, desktop app, CLI workflow, paid token plan, and an ambitious developer promise. But the real paid-user experience breaks at the trust layer.
After installing MiniMax Code 3.0 on Windows and logging in successfully, the desktop app opens briefly and then crashes immediately. The CLI experience is also extremely slow, so there is no reliable fallback path.
This is not just a QA issue. It is a product vision, product experience, and commercial readiness issue.
Product / plan
- Product: MiniMax Code / MiniMax Agent desktop app
- Version: MiniMax Code 3.0.x / latest Windows installer
- Platform: Windows
- Plan: Paid Token/Plus plan, approximately $20/month
- CLI: MiniMax Code CLI / MiniMax CLI workflow also tested
Actual behavior
- Install MiniMax Code 3.0 on Windows.
- Open the desktop app.
- Log in successfully.
- The app opens briefly.
- The app crashes immediately / closes instantly.
The desktop app cannot be used.
Separately, the CLI experience is extremely slow and does not feel acceptable for a paid coding-agent product.
Expected behavior
A serious paid AI coding product should provide:
- Reliable install and first-run experience
- Stable login and session handling
- Fast first successful action
- Responsive CLI fallback path
- Clear error states
- Discoverable logs and crash diagnostics
- Obvious recovery instructions
- Clear support and refund path for paid users
If authentication, gateway, GPU/WebView, daemon startup, or service initialization fails, the product should explain what failed instead of silently crashing.
Related reports already visible in this repo
This does not look isolated:
Impact
Severity: critical / blocking
For a paid user, the activation experience is currently broken:
- Desktop app crashes immediately after login.
- CLI feels unusably slow.
- There is no clear recovery path.
- There is no obvious diagnostic path for normal users.
- The paid value proposition does not match the delivered experience.
This is where customer trust is lost.
Product concern
QA catches defects.
Product leadership defines the standard of experience a company is willing to put in front of paying customers.
For the US developer market, this level of product experience does not feel commercially ready. A product is not judged by its launch page or model capability alone. It is judged by the first 10 minutes of real customer experience.
A serious AI product needs more than a capable model. It needs a complete trust experience around the model: positioning, activation, reliability, latency expectations, failure handling, diagnostics, support, and refund clarity.
Without that, users do not think, “This model is powerful.” They think, “This product is not ready.”
Request
Please provide one of the following:
- A working fix or workaround for Windows users affected by immediate post-login crashes.
- Clear instructions for collecting logs / crash dumps, including exact Windows paths.
- A rollback or stable installer if recent 3.0.x builds are unstable.
- A clear refund/support path for paid users whose desktop app and CLI are not usable.
======
Additional performance concern
Part of the reason this is disappointing is that I had already tested the model through an Ollama cloud service workflow. The model itself was acceptable, but the speed was slow.
My expectation was that the official MiniMax paid experience — desktop app, CLI, and direct MiniMax infrastructure — would provide a meaningfully better speed and reliability profile than using a third-party Ollama cloud service.
That did not happen.
In my experience, the official MiniMax CLI/app workflow feels extremely slow as well, and in some cases even slower than the Ollama cloud service workflow I had already tested.
This creates a major product value problem:
If the official paid product is not faster, not more stable, and not easier to recover from than an indirect cloud workflow, then the commercial value proposition becomes unclear.
For a paid AI coding tool, users expect the official/direct experience to provide at least one of the following:
- Better speed
- Better reliability
- Better UX
- Better diagnostics
- Better recovery
- Better support
- Better overall productivity
Right now, my experience is the opposite: the product is slower than expected, the desktop app crashes after login, and the fallback CLI path does not feel competitive.
Summary
This should be treated as more than a Windows crash bug.
MiniMax Code 3.0 may look strong on paper: AI coding agent, desktop app, CLI workflow, paid token plan, and an ambitious developer promise. But the real paid-user experience breaks at the trust layer.
After installing MiniMax Code 3.0 on Windows and logging in successfully, the desktop app opens briefly and then crashes immediately. The CLI experience is also extremely slow, so there is no reliable fallback path.
This is not just a QA issue. It is a product vision, product experience, and commercial readiness issue.
Product / plan
Actual behavior
The desktop app cannot be used.
Separately, the CLI experience is extremely slow and does not feel acceptable for a paid coding-agent product.
Expected behavior
A serious paid AI coding product should provide:
If authentication, gateway, GPU/WebView, daemon startup, or service initialization fails, the product should explain what failed instead of silently crashing.
Related reports already visible in this repo
This does not look isolated:
Impact
Severity: critical / blocking
For a paid user, the activation experience is currently broken:
This is where customer trust is lost.
Product concern
QA catches defects.
Product leadership defines the standard of experience a company is willing to put in front of paying customers.
For the US developer market, this level of product experience does not feel commercially ready. A product is not judged by its launch page or model capability alone. It is judged by the first 10 minutes of real customer experience.
A serious AI product needs more than a capable model. It needs a complete trust experience around the model: positioning, activation, reliability, latency expectations, failure handling, diagnostics, support, and refund clarity.
Without that, users do not think, “This model is powerful.” They think, “This product is not ready.”
Request
Please provide one of the following:
======
Additional performance concern
Part of the reason this is disappointing is that I had already tested the model through an Ollama cloud service workflow. The model itself was acceptable, but the speed was slow.
My expectation was that the official MiniMax paid experience — desktop app, CLI, and direct MiniMax infrastructure — would provide a meaningfully better speed and reliability profile than using a third-party Ollama cloud service.
That did not happen.
In my experience, the official MiniMax CLI/app workflow feels extremely slow as well, and in some cases even slower than the Ollama cloud service workflow I had already tested.
This creates a major product value problem:
If the official paid product is not faster, not more stable, and not easier to recover from than an indirect cloud workflow, then the commercial value proposition becomes unclear.
For a paid AI coding tool, users expect the official/direct experience to provide at least one of the following:
Right now, my experience is the opposite: the product is slower than expected, the desktop app crashes after login, and the fallback CLI path does not feel competitive.