Grok 4.7 is the latest push from SpaceXAI into serious coding, long-running agents and professional knowledge work — with the company emphasizing self-checking, longer context and aggressive API pricing.
SpaceXAI says Grok 4.7 is its most capable model for coding and knowledge work, using a larger base model and training that places more weight on difficult tasks that can take hours rather than minutes.
What changed in Grok 4.7
SpaceXAI announced Grok 4.7 on 21 September 2026. According to the company, the model was trained with a longer reinforcement-learning run on a harder mix of tasks and is better at checking its own work, handling longer context and operating inside the Grok Bot agent environment.
The direction is important. Frontier AI is moving beyond a model answering a prompt in one shot. The emerging battleground is whether an AI agent can keep working through a larger task, use tools correctly, recover from mistakes and produce something that is actually usable at the end.
Coding is a major focus
SpaceXAI reports gains over Grok 4.6 across several software-engineering and professional-work evaluations. On CursorBench 4.0, which stresses ambiguous and longer-running coding work, the company reports a 46.3% score for Grok 4.7 at its highest listed effort setting, compared with 40.4% for Grok 4.6.
The company also highlights improvements in document and presentation creation, electrical engineering, legal work and longer terminal-based tasks. These are vendor-reported benchmark results and should not be read as a universal ranking: performance can change significantly depending on the tools, prompts, data and guardrails around the model.
Pricing makes the competition more interesting
At launch, Grok 4.7 was priced from US$2 per million input tokens and US$6 per million output tokens. SpaceXAI also announced a faster variant at a higher price.
That puts pressure on the rest of the market because AI buying decisions are increasingly about the full cost of completing a task — not simply the token price or the top benchmark result. A model that finishes a workflow in fewer steps, uses fewer tokens or needs less human correction can be cheaper even when its headline API price is higher, and the reverse is also true.
What Grok 4.7 means for business automation
For companies considering autonomous or semi-autonomous agents, improvements in long-horizon work are more relevant than conversational novelty. Examples include an AI agent that researches a prospect, prepares a briefing, updates a CRM, drafts follow-up communication and flags uncertain information for a human.
SpaceXAI is also tying Grok 4.7 closely to Grok Bot, its persistent-agent environment. That is part of a broader industry shift toward AI systems that can carry responsibility over time rather than waiting for a new prompt for every action.
Grok’s progress is another reason businesses should avoid designing an automation around a single model vendor. A strong architecture separates the workflow, business rules and integrations from the model layer. That makes it easier to compare GPT, Grok, Claude and future models on quality, speed and cost without rebuilding the entire process.
Safety and availability
SpaceXAI says Grok 4.7 uses a new safeguard stack and reports stronger jailbreak resistance while retaining usefulness for legitimate cybersecurity work. At launch, the model was available through the Grok API, Grok Build, Cursor and other supported platforms.
