Grok 4.6
Grok 4.6 is xAI's flagship multimodal reasoning and agentic model engineered for complex problem-solving and autonomous project execution. Powered by advanced reinforcement learning and massive compute, it delivers frontier performance across agentic coding, multi-step logical reasoning, and interactive design workflows. With real-time X data synthesis, multi-step self-verification, and high-throughput execution, Grok 4.6 efficiently powers autonomous AI agent development, complex software engineering, academic research, and enterprise data analysis pipelines.
| Token Type | Price (USD) | Unit |
|---|---|---|
| Input | $2 | Per million tokens |
| Output | $6 | Per million tokens |
| Cache Read | $0.30 | Per million tokens |
Read Me
Grok 4.6
Grok 4.6 is xAI's flagship large language model released on August 12, 2026, designed for complex reasoning, programming, and agentic workflows, excelling at multi-step problem solving, command-line assistance, and high-quality software tasks. The model accepts text and image input with text output, a 500K token context window, and a knowledge cutoff of February 1, 2026. It supports four reasoning effort levels (low/medium/high/xhigh, default high), scoring 61 on the Artificial Analysis Intelligence Index (matching GPT-5.6 Sol Max), with DeepSWE v1.1 at 65.9% and CursorBench v3.2 at 69.9%.
The iCreat platform offers this model via the endpoint https://api.icreat.ai/v1/chat/completions (OpenAI Chat Completions compatible), at $2 per million input tokens, $6 per million output tokens, and $0.3 per million cached read tokens.
Model Positioning
Grok 4.6 is positioned as xAI's flagship model, designed for long-running agentic workflows, coding tasks, and interactive visual projects.
The model is a post-training upgrade over Grok 4.5 (not a full retraining), delivering significant improvements in long-running agents, programming knowledge work, and interactive visual projects through more extensive supplementary training, enhanced optimizers, and agentic reinforcement learning. It is recommended when output quality and reliability matter more than raw throughput.
On the iCreat platform, the model is served through an OpenAI Chat Completions compatible endpoint, and existing OpenAI SDKs work directly.
Core Capabilities
Complex Reasoning and Coding
Designed for complex reasoning, programming, and agentic workflows, with DeepSWE v1.1 at 65.9% and CursorBench v3.2 at 69.9%, excelling at multi-step problem solving and command-line assistance.
Long-Horizon Agentic Workflows
Supports continuous multi-step workflows (analysis → implementation → optimization → self-checking), excelling in long-horizon agent tasks. AA Intelligence Index score of 61, matching GPT-5.6 Sol Max.
Multimodal Input
Accepts text and image input with text output, handling interactive visual projects and image understanding tasks.
Four Reasoning Effort Levels
Supports low/medium/high/xhigh reasoning levels (default high), allowing developers to flexibly balance reasoning depth against cost and latency. xhigh is the maximum reasoning intensity.
Streaming Output and Function Calling
Supports streaming output (SSE), function calling, structured output, code execution, web search, and X search, covering full agentic workflow capabilities.
Pricing
| Token Type | Unit Price | Unit |
|---|---|---|
| Input | $2 | per million tokens |
| Output | $6 | per million tokens |
| Cached Read | $0.3 | per million tokens |
Note: The above are iCreat platform prices, consistent with xAI's official pricing (input $2, output $6, cache read $0.50 per million tokens). iCreat's cache read price is $0.3, lower than the official $0.50. The iCreat platform price shall prevail.
Application Scenarios
- Long-horizon agentic coding workflows (multi-step problem solving, command-line assistance)
- Complex software engineering and high-quality coding tasks
- Interactive visual projects and image understanding
- Enterprise API integration and real-time interaction scenarios
- Multi-step reasoning and knowledge workflows
Model Comparison
Comparison Table 1: Grok 4.6 vs Grok 4.5
| Feature | Grok 4.6 | Grok 4.5 |
|---|---|---|
| Positioning | Flagship, post-training upgrade | Previous flagship |
| Release Date | 2026-08-12 | 2026 |
| Context Window | 500K tokens | 500K tokens |
| AA Intelligence Index | 61 | 56 |
| DeepSWE v1.1 | 65.9% | 54.0% |
| CursorBench v3.2 | 69.9% | 66.7% |
| Reasoning Levels | low/medium/high/xhigh | low/medium/high |
| Input Price | $2/M tokens | $2/M tokens |
| Output Price | $6/M tokens | $6/M tokens |
Note: 4.6 is a post-training upgrade over 4.5, improving across agent coding and knowledge workflows at the same price, adding the xhigh reasoning level.
Comparison Table 2: Same-Tier Flagship Models
| Feature | Grok 4.6 | GPT-5.6 Sol | Claude Fable 5.1 |
|---|---|---|---|
| Positioning | xAI flagship, agent coding | OpenAI flagship | Anthropic's most capable public model |
| Context Window | 500K tokens | 1M tokens | 1M tokens |
| Input Price | $2/M tokens | $4/M tokens | $10/M tokens |
| Output Price | $6/M tokens | $20/M tokens | $50/M tokens |
| AA Intelligence Index | 61 | 61 | 65.7 |
| DeepSWE v1.1 | 65.9% | 73.0% | Per official |
| Reasoning Levels | low/medium/high/xhigh | Multiple | low/medium/high/xhigh/max |
Note: Grok 4.6 is the lowest-priced of the three ($2/$6), with a smaller context window (500K vs 1M); AA Intelligence Index matches GPT-5.6 Sol.
Why Choose Grok 4.6?
- xAI's flagship model, designed for long-horizon agentic coding and complex reasoning
- DeepSWE v1.1 at 65.9%, CursorBench v3.2 at 69.9%
- AA Intelligence Index score of 61, matching GPT-5.6 Sol Max
- Four reasoning effort levels (low/medium/high/xhigh) for flexible quality-cost balance
- Lowest price among comparable flagship models ($2/$6)
- Supports function calling, code execution, web search, and X search
Specifications
| Field | Value |
|---|---|
| Model Name | Grok 4.6 |
| Developer | xAI |
| Model ID | grok-4.6 |
| Endpoint | https://api.icreat.ai/v1/chat/completions |
| SDK base_url | https://api.icreat.ai/v1 |
| Release Date | 2026-08-12 |
| Model Type | Flagship LLM |
| Authentication | Authorization: Bearer |
| Context Window | 500,000 tokens (500K) |
| Input Modalities | Text, Image |
| Output Modalities | Text |
| Knowledge Cutoff | 2026-02-01 |
| Reasoning Levels | low / medium / high (default) / xhigh |
| Thinking Mode | thinking: {"type": "enabled"} |
| Streaming | Supported |
| Function Calling | Supported |
| Structured Output | Supported |
| Code Execution | Supported |
| Web Search | Supported |
| X Search | Supported |
| Prompt Caching | Supported |
| Billing Unit | Per million tokens |
Architecture
The iCreat platform's Grok 4.6 is served through an OpenAI Chat Completions compatible endpoint, allowing users to call it with any OpenAI API-compatible SDK. Requests must include the API Key in the Authorization header. The request body contains the model field (grok-4.6), a messages array (system/user/assistant messages), and optional thinking object (to enable thinking mode), reasoning_effort field (reasoning intensity), stream parameter (streaming output), and standard parameters such as temperature/max_tokens. The model is a post-training upgrade over Grok 4.5, supporting four reasoning levels (low/medium/high/xhigh) and delivering significant improvements in long-horizon agent and coding tasks. The model supports streaming, function calling, structured output, code execution, web search, and X search.
Notes
- The model field must use the iCreat platform's model_code (grok-4.6), not the vendor's original model name
- The model supports low/medium/high/xhigh reasoning levels, defaulting to high; xhigh is the maximum reasoning intensity
- The context window is 500K tokens, smaller than Claude (1M) and GPT-5.6 (1M)
- Cache read costs only $0.3/M tokens — 15% of the standard input rate, significantly reducing costs for agent scenarios
- The model is a post-training upgrade over Grok 4.5, not a full retraining
- Please safeguard your API Key and avoid hardcoding it in client-side code or public repositories
Frequently Asked Questions
How is the billing calculated?
Billing is based on actual token usage. Input is $2/M tokens, output is $6/M tokens, and cached read is $0.3/M tokens. For example, 1M input + 0.5M output costs $2 + $3 = $5. The iCreat platform price shall prevail.
What is the difference between Grok 4.6 and Grok 4.5?
4.6 is a post-training upgrade over 4.5 (not a full retraining), improving across AA Intelligence Index (61 vs 56), DeepSWE v1.1 (65.9% vs 54.0%), and CursorBench v3.2 (69.9% vs 66.7%) at the same price, adding the xhigh reasoning level.
What input modalities are supported?
Text and image input with text output. The model handles interactive visual projects and image understanding tasks.
How do I control reasoning effort?
Four levels: low, medium, high (default), and xhigh. Lower levels reduce cost and latency; higher levels improve reasoning depth. xhigh is the maximum reasoning intensity.
How large is the context window?
Supports 500K tokens (500,000 tokens) context window with a knowledge cutoff of February 1, 2026. The context window is smaller than Claude (1M) and GPT-5.6 (1M).
Is function calling supported?
Yes. The model is compatible with the OpenAI Function Calling protocol, and also supports structured output (JSON Schema), code execution, web search, and X search, covering full agentic workflow capabilities.