SpaceXAI has released Grok 4.6, the successor to the high-profile large language model the company previously shipped under the xAI brand. According to benchmarks published by the third-party evaluation firm Artificial Analysis, the new release matches OpenAI’s GPT-5.6 Sol at its maximum reasoning setting on the Artificial Analysis Intelligence Index, posting a score of 61. The result positions Grok 4.6 squarely in the top tier of general-purpose foundation models now available to enterprise developers and consumer applications.
The launch comes as competition among frontier-model laboratories has intensified, with Anthropic, OpenAI, and a growing number of challengers each releasing model families capable of multi-hour reasoning, agentic task execution, and software engineering assistance. SpaceXAI, which rebranded from xAI last month following its acquisition of the AI coding tool Cursor, is positioning Grok 4.6 as a system built for sustained agent workloads rather than single-turn question answering. Company materials describe the model as suitable for long-running agents, software development, knowledge work, and more ambitious visual projects.
Benchmark Performance and Competitive Positioning
On the Artificial Analysis Intelligence Index, Grok 4.6 scored 61, equal to GPT-5.6 Sol running at its highest reasoning effort and one point behind Anthropic’s Claude Fable 5 Max. The narrow margin suggests the three vendors are operating within a tight performance band at the upper end of current public benchmarks, with differentiations now emerging around price, latency, context length, and integration options rather than raw capability.
Artificial Analysis, which maintains one of the more closely watched independent leaderboards in the industry, weights scores from a range of evaluations covering reasoning, mathematics, and code. A score of 61 places Grok 4.6 ahead of every previously released SpaceXAI model and ahead of several prior OpenAI checkpoints, including earlier GPT-5 variants measured at default settings. The benchmark does not measure safety, hallucination rates, or instruction adherence, areas where independent reviewers have noted ongoing variance among top-tier models.
Distribution Channels and API Pricing
Grok 4.6 is available immediately through several distribution channels, including Cursor, Grok Build, the SpaceXAI application programming interface, OpenRouter, Vercel, and Cloudflare’s developer platform. The broad availability through inference providers and software development environments reflects a strategy designed to meet developers where they already work, rather than requiring them to integrate new tooling.
API pricing begins at $2 per million input tokens and $6 per million output tokens at the standard tier. A faster version, marketed toward latency-sensitive applications, costs twice as much. To encourage early adoption, Cursor and Grok Build users are receiving double their usual included Grok 4.6 usage during the model’s first week of availability.
- Intelligence Index score of 61 on Artificial Analysis, matching GPT-5.6 Sol maximum reasoning
- Available through Cursor, Grok Build, SpaceXAI API, OpenRouter, Vercel, and Cloudflare
- Standard API pricing at $2 per million input tokens and $6 per million output tokens
- Faster inference tier priced at double the standard rate during launch
- Launch-week promotional usage of twice the normal allowance for Cursor and Grok Build subscribers
Underlying Training Improvements
SpaceXAI attributed the benchmark gains to three principal training changes. The first was a longer supplemental training run, extending the compute applied to the underlying model after the initial pretraining phase. The second was a broader and higher-quality corpus of engineering data, drawn from public repositories and licensed sources. The third was an expanded reinforcement learning phase targeting coding tasks and structured knowledge work, areas where earlier Grok models had shown uneven performance relative to competitors.
Industry observers have noted that improvements in coding benchmarks have become a particular focus for frontier-model developers, as software engineering agents have emerged as one of the clearest near-term commercial applications for large language models. The expanded reinforcement learning phase, combined with richer engineering data, may help explain why SpaceXAI chose to ship Grok 4.6 in tandem with Cursor, the AI-assisted code editor acquired in June.
Companion Product: Grok Bot
Alongside the API release, SpaceXAI and Cursor have followed up on the acquisition with Grok Bot, a lightweight consumer-facing assistant available on iPhone, Mac, and additional platforms announced this week. The product is intended to extend Grok 4.6 beyond the developer console and into everyday workflows, though SpaceXAI has not disclosed detailed usage limits or subscription terms for the consumer version.
The combined release illustrates a pattern increasingly common among AI laboratories: bundling a model upgrade with a consumer-facing product launch, supported by an aggressive promotional window. Whether the launch-week usage bonus is enough to shift established developer preferences toward Grok 4.6 remains to be seen, particularly given OpenAI’s entrenched presence in enterprise procurement cycles and Anthropic’s reputation among software engineering teams. With Grok 4.6 now broadly available and benchmark parity with the leading models established, attention will shift to real-world evaluations of latency, cost per task, and reliability on long-horizon agent workflows as customers begin integrating the model into production systems.