MINIMAX_API_KEY and a group_id — missing either causes silent failures that look like network errors rather than auth errors.A 1 million token context window is not a marketing number. It means you can hand your OpenClaw agent an entire 800-page legal contract, a full codebase with 400 files, or three months of customer support transcripts — and process them in a single API call without chunking logic, without retrieval systems, without losing coherence across document boundaries. That is what MiniMax-Text-01 actually offers. And at early 2025 pricing, it costs significantly less per token than Claude for long-context tasks. Here is the setup that unlocks it.
What MiniMax Is
MiniMax is a Shanghai-based AI research company founded in 2021. They are not as well-known in Western markets as Anthropic or OpenAI, but their models compete directly on specific capabilities — particularly context length and multimodal processing.
Their global API is available at platform.minimaxi.com. Despite being a Chinese company, international access works reliably and the API design follows OpenAI-compatible patterns. The primary reason Western builders overlook MiniMax is the onboarding process, which has more steps than other providers, and sparse English-language documentation. Both are solvable problems that this guide addresses directly.
MiniMax operates two primary model lines relevant to OpenClaw integration:
- MiniMax-Text-01 — text generation with 1M token context window, the flagship model
- MiniMax-VL-01 — vision-language model for image and video understanding tasks
API Access Setup
The registration process is more involved than other providers. Work through these steps in order:
- Go to platform.minimaxi.com and create an account. International email addresses work.
- Complete email verification. The verification email sometimes lands in spam — check there first if it does not arrive within two minutes.
- Navigate to Account Settings → Group ID. Copy this numeric ID. You will need it in your config — this is different from every other provider and the step most people miss.
- Navigate to API Keys and generate a new key. Copy it immediately.
Set your credentials in the environment:
# .env file
MINIMAX_API_KEY=your-minimax-api-key-here
MINIMAX_GROUP_ID=1234567890123456789The Group ID is a long numeric string — typically 18–19 digits. It is not a slug or a name. Get the exact value from your account settings page.
MiniMax-Text-01 vs MiniMax-VL-01
| Property | MiniMax-Text-01 | MiniMax-VL-01 |
|---|---|---|
| Context window | 1,000,000 tokens | ~128,000 tokens |
| Vision input | No | Yes (images + video frames) |
| Text generation | Excellent | Good |
| Relative cost (input) | Lower | Higher |
| Best for | Ultra-long documents, analysis | Image Q&A, video understanding |
| Tool calling | Yes | Limited |
For most OpenClaw builders, MiniMax-Text-01 is the model you want. The 1M context window is the reason you chose MiniMax. VL-01 is the right choice when your specific agent task involves processing images or video frames alongside text — product image analysis, document scanning with embedded charts, or video content understanding.
openclaw.yaml Configuration
The MiniMax config has one field not present in other provider configurations: group_id. This is required for every API call and cannot be omitted.
model:
provider: minimax
model_id: MiniMax-Text-01
base_url: https://api.minimaxi.chat/v1
api_key_env: MINIMAX_API_KEY
group_id_env: MINIMAX_GROUP_ID
max_tokens: 8192
temperature: 0.7For vision tasks with MiniMax-VL-01:
model:
provider: minimax
model_id: MiniMax-VL-01
base_url: https://api.minimaxi.chat/v1
api_key_env: MINIMAX_API_KEY
group_id_env: MINIMAX_GROUP_ID
max_tokens: 4096
temperature: 0.5
vision: truehttps://api.minimaxi.chat/v1 — note the double-i in "minimaxi". Using https://api.minimax.chat/v1 (single i) routes to a different regional endpoint and will return inconsistent results or 404 errors for international accounts.Use Cases Where MiniMax Outperforms the Defaults
MiniMax is not a general-purpose replacement for Claude. It wins on a specific category of tasks.
Ultra-Long Document Analysis
Legal contracts, technical specifications, regulatory filings, multi-year financial reports. Any document exceeding 100K tokens becomes expensive and complex with Claude due to chunking overhead. MiniMax handles it in one call, preserving cross-document context that chunking destroys.
Codebase Understanding
Sending an entire repository to the model at once — rather than retrieval-augmented generation with partial context — produces dramatically better results for architecture questions, dependency tracing, and refactoring suggestions. A 400-file codebase comfortably fits in the 1M window.
Extended Conversation History
Agent pipelines that accumulate months of conversation context without truncation. Customer support agents that genuinely remember every previous interaction. Research assistants that maintain full session history across dozens of sessions.
Video Content Understanding
With MiniMax-VL-01, extract frames from video content and process them alongside transcripts. For agents handling video content moderation, video summarization, or scene analysis, the multimodal capability unlocks tasks that text-only models cannot handle.
Cost Comparison vs Claude for Long-Context Tasks
The cost advantage of MiniMax compounds as context length increases. These are approximate figures based on early 2025 pricing for input token processing:
| Document Length | Approx Tokens | Claude Sonnet Cost | MiniMax-Text-01 Cost | Savings |
|---|---|---|---|---|
| 50-page report | ~40K tokens | ~$0.12 | ~$0.05 | 58% |
| 200-page contract | ~160K tokens | ~$0.48 | ~$0.16 | 67% |
| 800-page filing | ~640K tokens | ~$1.92 | ~$0.64 | 67% |
| Full codebase | ~800K tokens | Not supported | ~$0.80 | — |
The last row is the real story: Claude Sonnet's context window tops out well below 1M tokens. For tasks that require the full window, MiniMax is not just cheaper — it is the only option through OpenClaw that can handle the task at all.
Limitations to Understand Before You Build
MiniMax's strengths are specific. Here is where it falls short:
- Response latency. Processing a 500K token context takes significantly longer than a short Claude call. For interactive agents where users wait for responses, the latency is noticeable. MiniMax is best suited for batch processing tasks that run in the background.
- English documentation gaps. The official docs are primarily in Chinese. English documentation exists but lags behind updates. Community resources in Western forums are sparse compared to Anthropic or OpenAI integrations.
- Dual-credential auth. The Group ID requirement adds complexity to deployment across multiple environments. You need to manage two credentials instead of one in your secrets management setup.
- Smaller tool ecosystem. OpenClaw's built-in tool registry has less testing coverage against MiniMax than against Claude or OpenAI. Tool-heavy agent patterns may require additional integration work.
- Quality variance at short context. For tasks under 50K tokens where Claude and MiniMax are similarly priced, Claude generally produces higher-quality outputs. Use MiniMax where its context advantage matters, not as a default replacement.
Frequently Asked Questions
What is MiniMax and where is the API hosted?
MiniMax is a Chinese AI company offering a global API at platform.minimaxi.com. Despite being China-based, they provide international accounts with global endpoint access. Their Text-01 model features a 1 million token context window and their VL-01 model adds vision capabilities for image and video analysis tasks.
How do I get a MINIMAX_API_KEY for OpenClaw?
Register at platform.minimaxi.com, verify your account, and navigate to API Keys to generate a key. You also need your Group ID from account settings — MiniMax authentication requires both values, unlike most other providers. The Group ID is a numeric string, not a name, and must be included in every API request through OpenClaw.
What is the context window size for MiniMax-Text-01?
MiniMax-Text-01 supports a 1 million token context window as of early 2025 — the largest available through OpenClaw. This makes it the definitive choice for processing entire codebases, long legal documents, or multi-session conversation histories in a single API call without chunking logic.
Does MiniMax-VL-01 support video understanding in OpenClaw?
MiniMax-VL-01 supports image input and video frame analysis. Full streaming video understanding depends on how frames are extracted and passed to the model. For video content, extract key frames and send them as a multi-image input — the model's context window accommodates substantial frame batches alongside text prompts.
How does MiniMax cost compare to Claude for long-context tasks?
For tasks requiring 200K+ tokens of context, MiniMax-Text-01 is significantly cheaper than Claude. Processing a 500-page document with MiniMax costs roughly 60–70% less than the equivalent Claude Sonnet call. The cost advantage compounds as context length increases, and for tasks requiring more than Claude's maximum context, MiniMax is the only option.
What are the limitations of using MiniMax in OpenClaw?
MiniMax's primary limitations are response latency (slower than Claude for short tasks), primarily Chinese market support with limited English documentation, and the dual-credential authentication requirement that catches most new users off guard. For tasks under 100K tokens where speed matters, Claude is typically the stronger choice.
Can I use MiniMax for voice or TTS tasks through OpenClaw?
Yes. MiniMax offers a Text-to-Audio (T2A) API accessible through OpenClaw's tool interface. The T2A model produces high-quality speech synthesis and supports multiple voices. Configure it as a separate tool call in your agent pipeline rather than the primary model endpoint — it operates independently from the Text-01 and VL-01 models.
Why does my MiniMax API call fail with a 401 error?
A 401 from MiniMax means the API key or Group ID is invalid or missing. Unlike other providers, MiniMax requires both credentials in every request. Check that your MINIMAX_API_KEY is set correctly and that the group_id in your openclaw.yaml matches your account's Group ID exactly — it is a long numeric value, not a text slug, and even one wrong digit fails authentication.