Grok Build, SpaceXAI's AI coding agent, is now accessible only through the SuperGrok Heavy plan at $300 per month for consumers, but developers should be cautious about the benchmark claims surrounding it. The widely circulated 70.8% SWE-bench Verified score attached to Grok Build across dozens of articles actually belongs to grok-code-fast-1, a different model that was retired in August 2026. Independent testing by Vals AI measured that older model at 57.6% rather than 70.8%, revealing a significant gap between marketing claims and verified performance.

Why Are Grok Build Benchmarks So Unreliable?

Neither OpenAI nor SpaceXAI published SWE-bench Verified scores for their current flagship models, making direct performance comparisons nearly impossible. OpenAI stopped reporting the benchmark after finding flawed test cases in a majority of its hardest unsolved problems. This creates a credibility vacuum that marketing claims rush to fill. Any article claiming to provide head-to-head SWE-bench numbers for Grok 4.6 and GPT-6 Astra is likely quoting older models or presenting unverified data.

The 70.8% figure has become so entrenched in coverage that it functions as accepted fact, even though it measures a tool that no longer exists. This matters because developers evaluating whether to spend $300 monthly on Grok Build are making decisions based on inflated performance claims. The actual performance gap between the retired model's verified 57.6% score and its advertised 70.8% represents a 22% overstatement of capability.

What Changed in SpaceXAI's Pricing Structure?

SpaceXAI restructured its subscription model in June 2026, replacing per-product daily caps with a single weekly usage pool shared across Chat, Imagine, Voice, and Build. The company has not published the exact size of this pool for any tier, leaving developers uncertain about actual usage limits. This opacity makes it difficult to calculate the true cost of using Grok Build beyond the base $300 subscription.

The $300 SuperGrok Heavy plan is now the only consumer tier that unlocks both Grok Bot and the full eight-agent version of Grok Build. This pricing strategy reflects SpaceXAI's positioning of Grok Build as a premium tool following its acquisition of Cursor, the popular AI code editor, in a $60 billion all-stock deal that closed in August 2026.

How Does Grok Build's Pricing Compare to Competitors?

The pricing gap between Grok Build and alternatives is substantial. ChatGPT's $20 per month Plus plan includes Codex, OpenAI's coding agent, alongside GPT-5.6 Sol, the company's flagship model. For developers on tighter budgets, the difference between $20 and $300 monthly represents a 15-fold price increase for access to Grok Build's advanced capabilities.

• ChatGPT Plus ($20/month): Includes Codex coding agent, GPT-5.6 Sol flagship model, ChatGPT Work, and Sites, with no usage pool restrictions published.
• SuperGrok ($30/month): Provides DeepSearch, voice mode, and full Imagine capabilities, but does not unlock the complete Grok Build feature set or multi-agent support.
• SuperGrok Heavy ($300/month): The only consumer plan offering multi-agent support and the full eight-agent version of Grok Build, along with Grok Bot access.

How to Evaluate Grok Build Before Committing to the $300 Investment

• Ignore the 70.8% benchmark claim: When evaluating Grok Build's coding performance, disregard the widely advertised 70.8% SWE-bench figure, which belongs to a retired model. Look instead for recent, independently verified benchmarks from sources like Vals AI, and recognize that neither SpaceXAI nor OpenAI currently publishes SWE-bench Verified scores for their flagship models.
• Test on your actual code: Rather than relying on published benchmarks, run Grok Build against your own codebase or representative coding tasks. The joint training of Grok 4.5 with Cursor on real developer session data may provide advantages for specific workflows that benchmarks don't capture.
• Calculate your true usage costs: Before subscribing, understand that SpaceXAI's weekly usage pool system means your actual costs depend on pool sizes the company has not published. Contact SpaceXAI support to clarify usage limits for the SuperGrok Heavy tier before committing.
• Compare against lower-cost alternatives: ChatGPT's $20 Plus plan includes Codex, which provides coding agent access at a fraction of Grok Build's cost. Evaluate both tools' actual performance on your specific coding tasks rather than relying on published benchmarks.

What Does This Pricing Strategy Signal About SpaceXAI's Direction?

SpaceXAI's decision to gate Grok Build behind a $300 monthly subscription signals confidence in the tool's capabilities but also creates clear market segmentation. Developers seeking affordable AI coding assistance have alternatives at lower price points, while those willing to pay premium prices gain access to multi-agent workflows and Grok Bot. The lack of published usage pool sizes for each tier adds uncertainty to the actual cost of using these tools beyond the base subscription fee.

The broader context matters here. SpaceXAI released Grok 4.6 on August 12, 2026, just weeks before OpenAI released GPT-6 Astra on September 3, 2026. Both companies are rapidly iterating on their models and pricing structures. Elon Musk has indicated that Grok 4.7 is close to release, though it had not shipped as of September 9, 2026. Developers should monitor these releases closely, as new model versions may shift the competitive landscape and pricing dynamics.

For teams evaluating coding agents, the key takeaway is that Grok Build's premium pricing reflects SpaceXAI's positioning of the tool as a high-end offering. Developers on limited budgets should carefully compare the actual performance of Grok Build against lower-cost alternatives before committing to the $300 monthly investment. The gap between the 70.8% marketing claim and the 57.6% verified performance of the retired model should serve as a reminder to verify claims independently rather than accepting published benchmarks at face value.