xAI’s most capable model, trained on longer multi-hour tasks and pitched at coding and knowledge work. It has a 500K tokens context window. Through xAI’s API it costs $2.00 (<200K) · $4.00 (≥200K) per million input tokens and $6.00 (<200K) · $12.00 (≥200K) per million output tokens.
Among the tools we track, it’s available in Cursor, Copilot, Kiro, Amp and Zed. Cursor says it co-trained the model with xAI, and offers a faster variant at twice the price. Zed made it the default for its xAI providers.
What xAI reports
- xAI reports 71.0% on DeepSWE v1.1 (high effort)
- xAI reports 37.6% on Terminal-Bench 4.0 (xHigh)
Vendor-reported figures, not independently verified by Code×Radar.
Related changes
- Copilot will drop six models, including Claude Fable 5, on December 14 · November 13, 2026
- Copilot retires GPT-5.5, GPT-5.4, GPT-5 mini, Grok 4.5 and Gemini 3.7 Flash · October 19, 2026
- Zed 1.22 lets subagents use different models and adds GPT-6.1 Sol and Grok 4.7 · September 30, 2026
- Grok 4.7 lands in Cursor at Grok 4.6's price, with a 500k long-context mode · September 21, 2026
Sources
Facts last checked November 15, 2026.