xAI releases Grok 4.6 at Grok 4.5 base prices

xAI released Grok 4.6 on 12 August 2026. The launch post and docs frame it as a step up from Grok 4.5 for long-running agents, coding, and knowledge work, with stronger interactive and visual first passes, at the same base token rates as 4.5 under 200k prompt tokens.

From the SpaceXAI launch post: "Introducing Grok 4.6. It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price."

Key facts

• Official primary: Introducing Grok 4.6 (12 August 2026). API model id `grok-4.6` on the model card and release notes.
• Availability: Cursor, Grok Build, Grok Bot, xAI API, plus partners OpenRouter, Vercel, and Cloudflare. First-week 2x included usage in Cursor and Grok Build per the launch post.
• Pricing (docs): $2 input / $0.50 cached input / $6 output per 1M tokens below 200k prompt tokens; $4 / $1 / $12 above. A fast variant is twice the base price. Reasoning effort: low, medium, high (default), xhigh.
• Context and modalities: 500,000-token context; text and image input, text output; no text output limit on the docs card.
• Lab-reported highlight scores (launch table, High where labeled): Artificial Analysis Intelligence Index 61 (ties GPT-5.6 Sol Max; Fable 5 Max 62; Grok 4.5 High 56); GDPVal-AA v2 1753; CursorBench v3.2 69.9%. Other rows on the same table trail some competitors (for example DeepSWE and Terminal-Bench). Full grid is on the blog.
• Training note: longer supplemental run than 4.5, Grok 4.5-regenerated SFT trajectories, agentic RL across coding and knowledge-work environments. NVIDIA on X says the model runs and was trained on NVIDIA GB300 NVL72 with NVLink.

How to read

Use the SpaceXAI post as the alert and the x.ai news page plus docs card as the ship surface. Composite and coding scores are lab- or leaderboard-published figures, not independent Frontier Flux runs. Same-day social posts may add price-performance color or tease later versions; they are not the product card.

Independence

Frontier Flux is not affiliated with xAI, SpaceXAI, or NVIDIA. We link official posts and docs; we do not rank models or endorse vendor benchmarks.