Grok
SpaceXAI's models, strong on long-running engineering tasks.
Grok is SpaceXAI's model family, sold through its own API rather than published as weights. The recent versions are pitched at coding and agentic work — long software-engineering runs where the model checks its own output — and take text, images and files with a 500K-token context. Closed weights and paid only, so treat it as a frontier option you rent, not one you host. The Grok chat app is listed separately under Chats.
At a glance — Grok 4.20 Multi-Agent
- Cost
- $1.25 in · $2.50 out per 1M tokensMid-range
- Memory
- 2M tokens — about 3,000 pages
- Reads
- Text · Images · Files
- Can do
- Use tools (agents) (not supported)
- Structured output
- Thinking mode
- Reliability
- 98.04% uptime, last 24h
- Released
- Mar 31, 2026
Advanced details
- API model ID
x-ai/grok-4.20-multi-agent- Max output
- 1.8M tokens
- Cached input
- $0.20 per 1M
- Prompts over 200K
- $2.50 in · $5.00 out per 1M
- Knowledge cutoff
- 2025-09-01
- Tokenizer
- Grok
Providers
The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.
| Provider | Input /1M | Output /1M | Cached /1M | Max output | Uptime 24h |
|---|---|---|---|---|---|
| xAI · zdr | $1.25 | $2.50 | $0.20 | 1.8M | 87.61% |
| xAI | $1.25 | $2.50 | $0.20 | 1.8M | 78.42% |
| xAI · zdr | $2.50 | $5.00 | $0.40 | 1.8M | — |
| xAI · priority | $2.50 | $5.00 | $0.40 | 1.8M | 98.04% |
Supported parameters
- include_reasoning
- logprobs
- max_tokens
- reasoning
- reasoning_effort
- response_format
- seed
- structured_outputs
- temperature
- top_logprobs
- top_p
Live from OpenRouter, refreshed hourly.
Key info
- Pricing
- Paid
- Category
- Models
- Best for
- Intermediate
- Made by
- SpaceXAI
- Weights
- Closed, API only
- Good for
- Long software-engineering tasks