Grok 4.20
by xAI · Released July 2025
xAI's frontier reasoning model. Grok 4 integrates real-time information from X (formerly Twitter) with frontier-grade reasoning. Built with massive reinforcement learning on a 200K-GPU supercluster, it excels at complex problem-solving, coding, and real-time analysis.
Grok 4.20
Powered by xAI · Mixture-of-Experts Transformer (~1.7T params)
Context Window
256K
Parameters
~1.7T (MoE)
Max Output
16K
Category
LLM Chat
Overview
Grok 4.20 is xAI's frontier reasoning model, built on a massive ~1.7 trillion parameter Mixture-of-Experts architecture and trained on a 200K-GPU supercluster with extensive reinforcement learning. It represents one of the largest and most compute-intensive training runs in AI history, resulting in a model with exceptional reasoning, coding, and analytical capabilities.
What sets Grok apart is its real-time integration with X (formerly Twitter) data, giving it access to current events, trending topics, and real-time public discourse that other models lack. The 256K token context window is large enough for substantial document analysis, and the multimodal text+image input support enables visual reasoning tasks alongside text-based ones.
The model excels at complex problem-solving that benefits from both deep reasoning and current information — trend analysis, real-time research, coding with awareness of latest library versions, and analytical tasks that require up-to-the-minute context. The massive RL training gives it strong performance on mathematical reasoning and competitive programming benchmarks.
Pricing
| Metric | Price |
|---|---|
| Input /1M tokens | ₹270.0000 |
| Output /1M tokens | ₹800.0000 |
1 credit = ₹1 = $0.01 USD. Prices shown from provider; CallMissed passes through with ~35% markup.
Key Highlights
- Real-time access to X (Twitter) data
- Trained on 200K-GPU supercluster with massive RL
- Multimodal: text and image input
- Strong coding and mathematical reasoning
Benchmarks
| Benchmark | Score |
|---|---|
| MATH-500 | 96.2% |
| GPQA Diamond | 79.3% |
| HumanEval | 92.4% |
| MMLU-Pro | 86.8% |
| AIME 2025 | 78.5% |
Technical Details
- Architecture: ~1.7 trillion parameter Mixture-of-Experts
- Trained on 200K-GPU supercluster with massive reinforcement learning
- Context window: 256K tokens
- Real-time X (Twitter) data integration for current information
- Multimodal: text and image input support
- Extensive RL post-training for reasoning and problem-solving
- Available via xAI API and CallMissed unified gateway
Strengths
- Real-time X data integration — unique capability among frontier models
- ~1.7T MoE architecture trained with massive RL on 200K GPUs
- Strong mathematical reasoning and competitive programming performance
- Multimodal text+image input for visual reasoning tasks
- 256K context window for substantial document analysis
Limitations
- Real-time data access is limited to X platform — not general web search
- 256K context is smaller than 1M-context competitors (GPT-5.4, Gemini, Opus)
- Proprietary and closed-source — no self-hosting option
- Less established ecosystem compared to OpenAI and Anthropic
Use Cases
API Example
curl https://api.callmissed.com/v1/chat/completions \
-H "Authorization: Bearer cm_YOUR_KEY" \
-d '{"model": "x-ai/grok-4.20", "messages": [{"role": "user", "content": "What are the trending topics on X right now about AI?"}]}'Endpoint: POST /v1/chat/completions · Model ID: x-ai/grok-4.20