SpaceXAI debuts Grok 4.6, claiming performance on par with GPT-5.6 Sol and Fable 5

If you use AI for coding, research or complex tasks, Grok 4.6 could be another model to watch. SpaceXAI has introduced its latest model with a focus on long-running agentic workflows, software engineering, complex visual projects and knowledge-intensive tasks. The company says Grok 4.6 is designed to stay on track across multi-step processes, explore unfamiliar codebases and assess its own output while working.

Longer training aims to improve reasoning

According to SpaceXAI, Grok 4.6 underwent a longer training phase than Grok 4.5. The process included an upgraded optimiser, refined training techniques, engineering data and curated synthetic datasets focused on complex technical concepts and reasoning.

Must Read: Spotify to label AI-generated artists with ‘AI Persona’; What it is and how it works

During development, supervised fine-tuning (SFT) trajectories were regenerated across reasoning problems, agent environments, software engineering, STEM and knowledge work. Evaluator models then filtered weaker trajectories to create an SFT checkpoint aimed at improving the model's behaviour and performance.

Reinforcement learning and benchmark performance

Grok 4.6 was also trained across agentic reinforcement-learning environments covering knowledge tasks, software development, computer-aided design, web development and kernel optimisation. SpaceXAI says the extended post-training process helped Grok 4.6 improve across benchmarks including Terminal-Bench, CursorBench, APEX-Agents and FrontierCode.

Must Read: xAI Launches Grok Bot: An always-on AI agent app that completes tasks on user's behalf

SpaceXAI's benchmark table shows Grok 4.6 High scoring 61 on the AA Intelligence Index, matching GPT-5.6 Sol Max, while Fable 5 Max scored 62. Grok 4.6 also led GPT-5.6 Sol Max and Fable 5 Max on GDPVal-AA v2, scoring 1,753, and on AA-Briefcase, scoring 1,577. However, it trailed the competing models on several other tests, including Terminal-Bench v3.0, where it scored 26% against 34.6% for GPT-5.6 Sol Max and 34.1% for Fable 5 Max.

Grok 4.6 price and availability

Grok 4.6 is available through Grok Build, Cursor, the SpaceXAI API and partner platforms including Cloudflare, OpenRouter and Vercel. The base API pricing is $2 per million input tokens and $6 per million output tokens, while the high-speed variant costs twice as much. During launch week, Grok Build and Cursor subscribers will also receive double their usual Grok 4.6 usage limits.