Grok 4.6 is SpaceXAI's upgraded frontier model, launched August 12, 2026. It extends Grok 4.5 with longer supplementary training and agentic reinforcement learning, reaching 61 on the Artificial Analysis Intelligence Index — tied with GPT-5.6 Sol Max — while keeping pricing at a disruptive $2/$6 per million tokens. The release leans hard into long-running agents and idea-to-product workflows.
Parameters
Undisclosed
Context Window
TBA
License
Proprietary
Release Date
2026-08-12
Benchmark Performance
AA Intelligence Index
61.0
LMArena Elo
1753.0
HLE
—
ARC-AGI-2
—
SWE-bench Verified
—
GPQA Diamond
—
MMLU-Pro
—
LiveCodeBench
—
AIME 2025
—
MATH-500
—
API Pricing
Input Price (per 1M tokens)
$2
Output Price (per 1M tokens)
$6
Billing Mode: standard
Strengths
- •Matches the top frontier models on the Artificial Analysis Intelligence Index (61) at roughly half the price of GPT-5.6 Sol or Claude Fable 5
- •Strong long-horizon agent capability: self-testing, verification, and multi-step task execution for research, data analysis, and coding
- •Competitive on agentic coding and knowledge-work benchmarks (GDPVal-AA v2 #1, plus strong showings on CursorBench v3.2 and DeepSWE v1.1)
- •Wide availability across API, OpenRouter, Vercel, Cloudflare, Cursor, and Grok Build with first-week double quotas
Weaknesses
- •Still trails GPT-5.6 Sol and Claude Fable 5 on several hard coding benchmarks despite the agentic focus
- •Context window length not yet disclosed, leaving maximum task scope unclear
- •Like its predecessors, subject to SpaceXAI's content and safety policy constraints
Use Cases
- •Agentic coding assistants and long-horizon software engineering
- •Autonomous research and data-analysis agents
- •Prototype build-out from a product brief to a runnable app
Deep Analysis
Artificial Analysis Intelligence Index
61
Tied with GPT-5.6 Sol Max; up from Grok 4.5 High's 56
GDPVal-AA v2 Elo
1753
#1, ahead of GPT-5.6 Sol Max (1728) and Claude Fable 5 (1741)
Input / Output Price
$2 / $6 per 1M
Half the cost of most frontier APIs; high-speed tier at double
Context Window
TBA
Not yet disclosed by SpaceXAI
Release
Aug 12, 2026
Available via API, OpenRouter, Vercel, Cloudflare, Cursor, Grok Build
Strengths
- ・Matches the top frontier models on the Artificial Analysis Intelligence Index (61) at roughly half the price of GPT-5.6 Sol or Claude Fable 5
- ・Strong long-horizon agent capability: self-testing, verification, and multi-step task execution for research, data analysis, and coding
- ・Competitive on agentic coding and knowledge-work benchmarks (GDPVal-AA v2 #1, plus strong showings on CursorBench v3.2 and DeepSWE v1.1)
- ・Wide availability across API, OpenRouter, Vercel, Cloudflare, Cursor, and Grok Build with first-week double quotas
Weaknesses
- ・Still trails GPT-5.6 Sol and Claude Fable 5 on several hard coding benchmarks despite the agentic focus
- ・Context window length not yet disclosed, leaving maximum task scope unclear
- ・Like its predecessors, subject to SpaceXAI's content and safety policy constraints
Competitor Comparison
| Model | Price |
|---|---|
| Claude Fable 5 (max) | $10/$50 |
| GPT-5.6 Sol (max) | $5/$30 |
| Grok 4.5 (high) | $2/$6 |
Grok 4.6 is SpaceXAI's upgraded frontier model, launched August 12, 2026. It extends Grok 4.5 with longer supplementary training and agentic reinforcement learning, pushing its Artificial Analysis Intelligence Index to 61 — tied with GPT-5.6 Sol Max — while keeping pricing at a disruptive $2/$6 per million tokens. The release leans hard into long-running agents and idea-to-product workflows.
Sources
Analysis generated: 2026-08-13