← Back to API models

Grok 4.6

by SpaceXAI (formerly xAI) · new Opus-class flagship · released August 12, 2026

SpaceXAI's newest flagship, replacing Grok 4.5 at the same standard-tier price. Pricing is tiered: $2/$6 per 1M tokens under 200K, $4/$12 per 1M once a request crosses 200K.

Input <200K $2 / 1M
Output <200K $6 / 1M
Context 500K
≥200K tokens $4 / $12
SpaceXAI Console ↗ Updated August 17, 2026
§ API pricing

Per-token rates, with a tier break at 200K.

Input <200K
$2/1M tokens
Standard tier
  • Applies while the prompt stays under 200K tokens
  • Cached input at $0.50 / 1M (up from $0.30 on 4.5)
  • Same input rate as the retired Grok 4.5
Output <200K
$6/1M tokens
Standard tier
  • Includes reasoning tokens
  • Low/medium/high effort configurable
  • 3× the standard-tier input rate
≥200K tokens
$4/$12 per 1M
Long-context tier
  • Kicks in once the prompt reaches 200K tokens
  • Applies to the full request, not just the overage
  • Double the standard input and output rates
Context
500Ktokens
Window
  • Unchanged from Grok 4.5
  • Web search, X search, code exec
  • Function calling supported

Grok 4.6 bills $2 / 1M input and $6 / 1M output for prompts under 200K tokens — unchanged from Grok 4.5. Once a request reaches 200K tokens, the entire request is billed at $4 / 1M input and $12 / 1M output. Cached input rose to $0.50 / 1M at the standard tier, up from $0.30 on 4.5.

What's new in Grok 4.6

SpaceXAI (the company formerly known as xAI, rebranded July 6, 2026 after fully merging into SpaceX) released Grok 4.6 on August 12, 2026, replacing Grok 4.5 as the flagship at the same $2/$6 standard-tier price. Grok's product name and API are unaffected by the corporate rebrand. The headline pricing change from 4.5 is cached input, which rose from $0.30 to $0.50 per 1M at the standard tier — everything else on the rate card carries over unchanged.

The model keeps the 500K-token context window, configurable low/medium/high reasoning effort, function calling, web and X search, and code execution that defined 4.5. Pricing stays tiered: requests under 200K tokens bill at $2 input / $6 output per 1M, but once a prompt reaches 200K tokens the whole request bills at $4 input / $12 output.

Capabilities

Grok 4.6 is built for coding and agentic work: tool-use loops, code execution, and real-time retrieval from the web and X. Its token efficiency and standard-tier output price make it attractive for high-volume agent runs, while native X search keeps its edge on current events and social context. The tiered pricing means the model rewards keeping prompts under 200K tokens — worth engineering around if your workload sits near the boundary. Availability spans the SpaceXAI API console, Grok Build, Cursor, OpenRouter, Vercel, Cloudflare, and Microsoft Office add-ins.

Typical use cases

  • Coding agents and IDE integrations
  • Real-time research over the web and X/Twitter
  • High-volume agent loops where token efficiency matters
  • Function-calling and tool-use pipelines
  • Current-events analysis and social monitoring

Sibling and rival comparison

ModelInput / 1MOutput / 1MContext
Grok 4.6 (<200K)$2$6500K
Grok 4.6 (≥200K)$4$12500K
Grok 4.3$1.25$2.501M
Claude Opus 5$5$251M
GPT-5.6 Sol$5$301M
Gemini 3.1 Pro$2$121M

Grok 4.6 undercuts the other flagships sharply on standard-tier output price — half of Gemini 3.1 Pro and a fifth of GPT-5.6 Sol — while claiming Opus-class quality. For a cheaper reasoning workhorse under the flagship, Grok 4.3 is SpaceXAI's mid-tier option.

← See all Grok / SpaceXAI plans