Elon Musk says Grok 4.6 is “objectively number one” when intelligence, speed and cost are considered together. Its early results are impressive, but winning benchmarks is not the same as winning enterprise customers—or making money.
According to Artificial Analysis, Grok 4.6 scored 61 on its Intelligence Index, matching OpenAI’s GPT-5.6 Sol Max. It sits just behind Anthropic’s leading Claude models while competing closely with Chinese challengers such as Kimi K3 and Qwen3.8 Max.
More importantly, Grok is cheap. Its API costs $2 per million input tokens and $6 per million output tokens, substantially undercutting the most powerful models from OpenAI and Anthropic.
That makes Grok highly competitive, but not an undisputed winner. It performs particularly well on agentic knowledge work, yet trails Claude and OpenAI on several important coding benchmarks. There is still no single test capable of declaring one model the best across every category.
The bigger impact of Grok 4.6 may therefore be on pricing.
OpenAI has already reduced the price of GPT-5.6 Luna by 80% and Terra by 20%. Its flagship Sol pricing remains unchanged, but the direction is clear: advanced intelligence is becoming cheaper. Grok’s aggressive pricing increases the pressure on Anthropic and other providers to respond.
At the same time, the best Chinese open-weight models are no longer merely ultra-cheap alternatives. As Kimi, Qwen and others approach US frontier performance, their developers can charge more—even if they generally remain cheaper than the most expensive American models.
The intelligence-price gap is narrowing from both directions. US providers are lowering prices while Chinese developers are charging more as their models improve. This makes it increasingly difficult for any company to maintain a lasting advantage based on price alone.
Grok is doing the industry a favour by forcing everyone to deliver more intelligence for less money. But it may be doing so at a significant cost to itself.
SpaceX’s AI segment—which includes Grok, X and its infrastructure business—generated $3.4 billion in revenue during the first half of 2026 while recording a $3.7 billion operating loss. It also invested $23.6 billion in AI infrastructure during the same period.
These figures do not prove that individual Grok API calls lose money. Paid inference may be profitable when SpaceXAI’s infrastructure is fully utilised. But after including training, research, free consumer usage and infrastructure depreciation, the wider Grok operation is almost certainly loss-making.
This raises the possibility that Musk is deliberately subsidising Grok to gain market share and start an AI price war. There is no direct evidence that Grok is being sold below its immediate inference cost. However, SpaceXAI is clearly willing to tolerate substantial losses while distributing it aggressively through X, Cursor, Grok Build and GitHub Copilot.
The $60 billion acquisition of Cursor strengthens that distribution, but it may also be a bet on a product category that is changing extraordinarily quickly.
I have experienced that shift myself. As a hands-on agentic coder, I began with Github Co-pilot then Windsurf, which felt like the best options until around July 2025. When Windsurf stumbled, I moved to Cursor. Since then, however, tools such as OpenCode, Hermes, Claude Code and OpenAI Codex have changed how I work.
The centre of gravity is moving from AI-assisted IDEs towards autonomous coding agents. These agents can examine an entire repository, plan a task, edit multiple files, run tests and continue working with less human supervision. That is not to say the IDE is unimportant, but IDEs are increasingly interchangeable. Claude Code and Codex can be integrated into VS Code—and even used within Cursor itself. The IDE is becoming the interface around the agent rather than the main source of value.
Among the developers I know, Claude Code and Codex are increasingly becoming their primary coding tools. I know people who have moved away from Cursor, but few who regretted leaving and felt compelled to return. That is anecdotal, but it demonstrates how quickly developer loyalty can shift when a better agent appears.
Cursor was one of the hottest products in AI coding. It is no longer obvious that it owns the future of the category. Unless Cursor evolves quickly enough into a compelling agent-orchestration platform, SpaceXAI could be buying yesterday’s market leader at tomorrow’s price.
Then there is the Elon Musk factor.
Musk’s ability to move quickly, raise enormous amounts of capital and pursue ambitious technical projects is a genuine advantage. But his wider public behaviour creates a risk that cannot be measured by benchmarks.
Enterprise buyers care about vendor stability, predictability and reputation. They do not want a critical technology supplier to become a board-level controversy or brand risk. For many companies, Musk is not simply another technology CEO: he is a polarising political and cultural figure.
The experience of Twitter—now X—is the obvious warning. X remains highly influential, but advertisers pulled back sharply following Musk’s takeover amid concerns about content, brand safety and his public statements. Reuters reported that US advertising revenue declined by at least 55% year-on-year in each month following the acquisition.
That history matters because enterprise trust is difficult to rebuild. Low prices may attract developers, but they do not necessarily reassure chief information officers, legal departments or corporate boards. Once large customers consider a vendor unpredictable, winning them back can cost far more than the original discount.
The risk for Musk is that OpenAI and Anthropic can lower their prices too. If the price gap becomes small, enterprises may stay with the providers they already know and trust.
My view is that Grok 4.6 will win workloads before it wins the market. It may even succeed in pushing down frontier-model prices across the industry.
But if every leading model becomes cheaper, customers will choose based on performance, ecosystem, stability and trust. In that contest, Musk’s behaviour may be as important as Grok’s intelligence.
Grok may win the price war for everyone. Whether it can win enough enterprise customers to justify the billions being spent is a much harder question.
Leave a Comment