Grok 4.6

2026-08-12 · Publisher : xAI · Domain : Artificial Intelligence

SpaceXAI launches Grok 4.6, a model for long-running agents and visual or interactive work, at USD 2 input and USD 6 output per million tokens, with a 500k-token context.

SpaceXAI launches Grok 4.6, a model for long-running agents and visual or interactive work, at USD 2 input and USD 6 output per million tokens, with a 500k-token context. ### What changes - Better autonomy on long tasks and increased self-testing. - Additional training on curated synthetic data, a new optimizer, reinforcement learning (GPU kernels, web, CAD). - Fast variant at 2 times the price. - Arrival in GitHub Copilot (August 14), Amazon Bedrock and Microsoft Foundry (August 26). ### Specifications - API identifier: grok-4.6. - Context: 500k tokens. - Availability: Cursor, Grok Build, API, OpenRouter, Vercel, Cloudflare. ### Pricing | Item | Input | Output | |---|---|---| | Standard (under 200k tokens) | USD 2 | USD 6 | | Cache | USD 0.50 | - | | Tier of 200k tokens and above | USD 4 | USD 12 | ### Benchmarks | Benchmark | Score | |---|---| | AA Intelligence Index | 61 | | CursorBench v3.2 | 69.9% | | DeepSWE v1.1 | 65.9% | | FrontierCode v1.1 | 61.3% | | APEX-Agents | 57.5% | Scores published by the vendor. ### Key takeaways Rapid spread across cloud partners, useful for code agents already deployed.

Tags : LLM, xAI, Grok, Agents, SpaceXAI