#7SpaceXAI (xAI)

Grok 4.7

SpaceXAI's newest Grok model, tuned for coding and knowledge work, with better checking of its own work and longer tasks than Grok 4.6. The company says it is twice as fast at half the price of comparable models. A Fast variant costs double for double the output speed.

• 500K-token context; text and image input
• Reasoning effort low to xhigh (default high)
• Lab-reported CursorBench 4.0: 46.3% (Grok 4.6: 40.4%)
• Lab-reported DeepSWE v1.1: 71.0%

AccessGrok API (grok-4.7), Grok Build, Cursor, and third-party coding tools, model routers and cloud platforms.

Price$2.00 input / $6.00 output per 1M tokens (cached input $0.50); Fast variant at 2x price

Grok 4.6

Built on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work. It shipped first in Grok Build and Cursor and reached the major cloud platforms within two weeks. SpaceXAI reported a score of 61 on the Artificial Analysis Intelligence Index.

• 500K-token context; text and image input
• Higher price tier for requests above 200K tokens
• Lab-reported DeepSWE v1.1: 65.9%
• Lab-reported FrontierCode v1.1: 61.3%

AccessGrok Build and Cursor at launch; API via console.x.ai; OpenRouter, Vercel, Cloudflare. Later GitHub Copilot (14 Aug), Amazon Bedrock (19 Aug), Gemini Enterprise Agent Platform (21 Aug), Microsoft Foundry (26 Aug).

Price$2.00 input / $6.00 output per 1M tokens (cached input $0.50); Fast variant at double price

Grok Imagine Image 2.0

SpaceXAI's second-generation image model, focused on following detailed instructions, laying out dense visuals and rendering small text cleanly. It adds editing tools such as region-targeted 'magic wand' edits, smart resizing and background removal. SpaceXAI says it ranks second on Arena leaderboards for both text-to-image and image editing.

• Region-targeted edits, segmentation and background removal
• Multi-reference editing with up to 5 input images
• Smart resize to other aspect ratios
• Lab-reported #2 on Arena for text-to-image and image editing

Accessgrok.com/imagine and the Grok iOS and Android apps (as Quality Mode); API as grok-imagine-image-2.0.

Grok Voice Think Fast 2.0

SpaceXAI's speech-to-speech model for real-time voice agents, which reasons while it speaks so thinking adds no delay. It transcribes better in noisy conditions, fires tool calls sooner and uses fewer reasoning tokens than version 1.0. SpaceXAI reports a leading 82.9% on Artificial Analysis' speech-to-speech quality index.

• Audio in, audio out for real-time voice agents
• Reasons in parallel with speech output
• Uses 0.4x the reasoning tokens of version 1.0
• Lab-reported 82.9% on Artificial Analysis S2S Quality Index

AccessxAI Voice API as grok-voice-think-fast-2.0; the grok-voice-latest endpoint moved to it on 5 Aug 2026 (1.0 can be pinned); also on Vercel AI Gateway.

Grok 4.5

The first Grok model trained alongside coding-tool maker Cursor, positioned as SpaceXAI's strongest model for coding, agent tasks and knowledge work. It set the $2/$6 per-million-token price its successors kept. It launched on the API, Grok Build and Cursor first and reached the consumer Grok apps about two weeks later.

• 500K-token context; text and image input
• Reasoning effort low to xhigh
• Lab-reported SWE-Bench Pro: 64.7%
• Lab-reported Terminal Bench 2.1: 83.3%

AccessxAI API (grok-4.5), Grok Build, Cursor (all plans); EU API from 17 Jul 2026; Grok apps on iOS, Android, web and X from 22 Jul 2026.

Price$2.00 input / $6.00 output per 1M tokens

Grok Imagine Video 1.5

SpaceXAI's updated video model with generated sound, aimed at better motion, physics and audio sync at higher speed. It makes 6-second 720p clips in about 25 seconds, down from over 40. It follows the January 2026 Grok Imagine API, which first opened xAI video generation and editing to developers.

• Text- and image-to-video with generated audio
• 720p and higher; 6-second 720p clip in about 25 seconds
• Clearer, better-synced speech; fewer warps across a clip
• Reference inputs added 31 Jul 2026 ('Video 1.5 with References')

AccessAPI (generally available) as grok-imagine-video-1.5; grok.com/imagine and the Grok iOS and Android apps; a Video 1.5 Fast variant.

Grok 4.3

A reasoning-first Grok model with always-on, adjustable reasoning effort, aimed at tool use, instruction following and document-heavy enterprise work such as contract review and financial Q&A. It offers a 1M-token context at a lower price than the later Grok 4.5 to 4.7 models. No launch post for it appears on x.ai/news.

• 1M-token context; text and image input
• Reasoning effort none to xhigh (default low)
• Function calling and structured outputs
• Knowledge cutoff December 2025 (Oracle OCI docs)

AccessxAI API (grok-4.3); Vercel AI Gateway and OpenRouter; Oracle OCI (1 May 2026), Amazon Bedrock (15 Jun 2026), Google Cloud partner models. Beta for SuperGrok Heavy subscribers from 17 Apr 2026 (press).

Price$1.25 input / $2.50 output per 1M tokens (cached input $0.20)

Grok 4.20

A Grok 4 update built around a multi-agent mode, in which several agents work on a problem in parallel and debate before answering. xAI's docs stress speed, agentic tool calling, low hallucination rates and prompt adherence. It ran as a beta in the Grok apps from February 2026 before the API release in March.

• Grok 4.20 Multi-agent variant on the API
• 1M-token context per docs.x.ai (press reports 2M)
• Text and image input; function calling, structured outputs
• Batch API supported

AccessGrok apps (beta from Feb 2026, per press); xAI API as grok-4.20-0309-reasoning, plus a Grok 4.20 Multi-agent variant.

Price$1.25 input / $2.50 output per 1M tokens (cached input $0.20), as shown on the current docs.x.ai model page

Grok 4.1 Fast

A low-cost Grok model built for tool-calling agents, which xAI called its best tool-calling model. It comes in reasoning and non-reasoning versions with a 2M-token context and launched alongside xAI's Agent Tools API. xAI aimed it at production agents for uses such as customer support and finance.

• 2M-token context window
• Reasoning and non-reasoning modes
• Built for agentic tool calling and search
• Tool calls from $5 per 1,000 successful invocations

AccessxAI API as grok-4-1-fast-reasoning and grok-4-1-fast-non-reasoning, launched with the Agent Tools API.

Price$0.20 input / $0.50 output per 1M tokens (cached input $0.05)

Grok 4.1

A Grok 4 update for the consumer apps focused on creative writing, emotional intelligence and conversation, while keeping analytical performance. xAI reported its Thinking version at #1 on LMArena's text leaderboard at launch and fewer hallucinations on information-seeking prompts. In xAI's test, users preferred it 64.78% of the time over the previous model.

• Thinking and non-reasoning versions
• Lab-reported LMArena text Elo 1483 (Thinking), 1465 (non-reasoning)
• Improved EQ-Bench3 and Creative Writing v3 results, per xAI
• Lower hallucination rate on info-seeking prompts, per xAI

AccessAll users on grok.com, X, and the Grok iOS and Android apps.