SpaceXAI Ships Grok 4.6, Matching GPT-5.6 Sol at Half the API Price
News Summary
SpaceXAI, the artificial intelligence group operating under Elon Musk's SpaceX umbrella and formerly known as xAI, officially released its newest large language model, Grok 4.6, on Wednesday, August 12, 2026 (Eastern Time). The launch was announced through an official blog post on the company's site and confirmed across multiple technology outlets, positioning Grok 4.6 as a frontier-class model with a particular emphasis on long-running autonomous agents, software coding, and complex knowledge work.
What Was Announced
Grok 4.6 became available immediately at launch across several surfaces, including Grok Build, the coding-focused Grok Bot agent system, the third-party developer tool Cursor, and the xAI API. The company also made the model accessible through infrastructure partners such as OpenRouter, Vercel, and Cloudflare, broadening the number of ways developers can integrate it into existing products. According to the official announcement, Grok 4.6 is built to sustain complex, multi-step tasks over long sessions, whether that means researching a topic across many sources, analyzing large volumes of information, navigating a full software codebase, or turning a rough idea into a finished application or document.
Model Capabilities and Specifications
Grok 4.6 supports a 500,000-token context window and accepts both text and image inputs, though it currently returns text-only output. The model has no fixed limit on output length for a single response, which the company says is intended to support long-form technical writing and extended agentic workflows without truncation. Compared with its predecessor, Grok 4.5, the new version reflects updated supervised fine-tuning and reinforcement learning from human feedback, changes the company credits for measurable gains in coding, terminal-command execution, and general knowledge-work benchmarks.
Independent Benchmark Results
Third-party evaluator Artificial Analysis measured Grok 4.6 (high) at a score of 61 on its Intelligence Index, a composite benchmark spanning nine individual tests covering reasoning, mathematics, scientific knowledge, and coding ability. That score places Grok 4.6 in a near tie with GPT-5.6 Sol, slightly behind Claude Opus 5 at 63 and Claude Fable 5 at 62, and ahead of Kimi K3. On agentic coding tasks specifically, Artificial Analysis found that Grok 4.6 resolved evaluation tasks in roughly 53 conversational turns while consuming about 0.5 billion input tokens on average, a notably more efficient path than the roughly 103 turns and 2.0 billion input tokens recorded for Claude Opus 5 on comparable tasks.
Pricing and Availability
SpaceXAI priced Grok 4.6 access through its API starting at 2 US dollars per million input tokens and 6 US dollars per million output tokens, a rate the company and outside reporting describe as roughly half the cost of comparable frontier models from rival labs. Elon Musk publicly commented on the release on his social media platform, citing an internal 1753 Elo rating for the model and describing the launch positively. Analysts covering the release noted that the combination of near-frontier benchmark performance and lower per-token pricing could pressure competitors to adjust their own pricing structures.
Company Roadmap
Reporting following the release indicated that SpaceXAI is already preparing a larger successor model, tentatively referred to as Grok 4.7, expected to scale up to roughly 2.1 trillion parameters and to arrive within weeks of the Grok 4.6 launch. Coverage also pointed to an internal target of releasing a next-generation Grok 5 model before the end of 2026, suggesting the company intends to maintain a rapid release cadence through the rest of the year.
Broader Product Context
Alongside the core model release, SpaceXAI also introduced Grok Bot, an enterprise-oriented system designed to coordinate multiple AI agents across departments such as sales, finance, and engineering. The tool is built to let agents share context and hand off tasks to one another, a design aimed at automating multi-step business workflows rather than handling single, isolated requests. Industry observers framed Grok Bot as SpaceXAI's answer to similar enterprise agent products being developed by other major AI labs, reflecting a broader industry shift toward AI systems that can operate semi-autonomously across longer, more complex workflows.