⚡ Uncle Cat AI Radar
ModelsAgentsIndustry

SpaceXAI Releases Grok 4.6 for Long-Running Agents

Grok 4.6 targets coding and sustained agent work with a 500,000-token context window and pricing below several frontier rivals.

A model tuned for sustained work

SpaceXAI has released Grok 4.6, a new frontier model aimed at coding, knowledge work and agents that operate across long sequences of tool calls. The model is available through the company’s API and has already appeared in third-party products including Devin and Poe, giving the launch immediate reach beyond SpaceXAI’s own interfaces.

Grok 4.6 supports a 500,000-token context window and includes native function calling, code execution, web search and X search. SpaceXAI also emphasizes context compaction and predictable prompt caching—features intended to control the growing cost and instability of agents that repeatedly inspect repositories, execute code and revise their plans. Standard API pricing is listed at $2 per million input tokens and $6 per million output tokens, with a faster service tier costing more.

Early independent testing from Artificial Analysis gave Grok 4.6 a score of 61 on its Intelligence Index, placing it alongside OpenAI’s GPT-5.6 Sol in that composite test while costing less for the measured workloads. Its AA-Briefcase results also indicate a substantial improvement in agentic knowledge work over Grok 4.5. Those numbers remain benchmark results rather than guarantees: performance and total cost can change sharply with the agent harness, context length, caching behavior and number of retries.

Why it matters

The release reflects a broader change in frontier-model competition. Vendors are no longer selling only higher answer quality; they are competing on whether a model can remain useful and economical throughout a multi-hour workflow. By combining competitive benchmark performance, built-in tools and aggressive pricing, SpaceXAI is trying to become a default worker inside other companies’ agent products. Day-one integrations suggest model distribution is increasingly decided inside coding agents and workflow platforms, not solely in consumer chat applications.

Sources