xAI has released Grok 4.6, a new flagship large language model tuned for long-running agents, software engineering and complex visual tasks that the company says matches OpenAI's GPT-5.6 Sol Max on the composite Artificial Analysis Intelligence Index at roughly half the token cost.
What xAI shipped
Grok 4.6 became available on August 12, 2026, through Grok Build, Cursor, the xAI API and partner platforms including Cloudflare, OpenRouter and Vercel. Base pricing is $2 per million input tokens and $6 per million output tokens, with a high-speed variant at double the cost and cached input at $0.50 per million. During launch week, Grok Build and Cursor subscribers receive double their usual Grok 4.6 usage limits.
Benchmark scorecard
On xAI's published tables, Grok 4.6 High scores 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol Max, while Anthropic ally Fable 5 Max scored 62. Grok 4.6 also leads on GDPVal-AA v2 with 1,753 and on AA-Briefcase with 1,577. The model trails competitors on Terminal-Bench v3.0, however, where it scored 26% against 34.6% for GPT-5.6 Sol Max and 34.1% for Fable 5 Max. Artificial Analysis notes that Grok 4.6 completes agent tasks in roughly half the turns of rival models, a meaningful efficiency win for cost-sensitive workflows.
Agentic training pipeline
xAI says Grok 4.6 underwent a longer training run than Grok 4.5, with an upgraded optimiser and curated synthetic datasets focused on complex technical reasoning. Supervised fine-tuning trajectories were regenerated across reasoning problems, agent environments, software engineering, STEM and knowledge work, then filtered by evaluator models. Reinforcement-learning environments covered knowledge tasks, software development, computer-aided design, web development and kernel optimisation, feeding into gains on Terminal-Bench, CursorBench, APEX-Agents and FrontierCode.
Race for cheaper agents
The release lands a day after Elon Musk's group unveiled Grok Bot, an always-on agent app that runs tasks on a user's behalf, and it follows OpenAI's July release of GPT-5.6-Cyber for enterprise security workflows and Anthropic's move to acquire Decart for $6 billion to cut inference costs. It also comes as investors keep repricing agentic coding leaders such as Cognition at more than $40 billion. Grok 4.6's pitch to developers is straightforward: comparable frontier intelligence at a materially lower token price, delivered through the tools they already use.
Reporting based on coverage from Business Today, SiliconANGLE and xAI benchmarks.