Quick answer
xAI released Grok 4.7 on September 21, 2026. It is a larger base model than Grok 4.5 with an extended reinforcement-learning phase, a 500,000-token context window, and API pricing of $2 per million input tokens and $6 per million output tokens — a fifth of GPT-6 Astra's output price and less than a third of Opus 5.5's. xAI published its own benchmark results, including 71.0% on DeepSWE and 64.0% on EEBench; no independent results are available yet.
Grok has spent 2026 as the assistant with the strongest personality and the weakest enterprise story. Grok 4.7 is xAI's attempt to change the second half of that: a bigger model, a bigger context, and a price that makes it the obvious default for anyone whose main constraint is budget.
What changed
- A larger base model than 4.5 — xAI has not published parameter counts
- Extended reinforcement learning after pre-training, the technique behind most of this year's reasoning gains
- Context window of 500,000 tokens, up from 256K on Grok 4
- API pricing of $2 input / $6 output per million tokens
- Available to X Premium and SuperGrok subscribers in the Grok app and through the xAI API
The benchmarks — vendor-reported
xAI reports 71.0% on DeepSWE, a software-engineering benchmark, and 64.0% on EEBench, among others. Those are strong numbers for the price, and behind the leaders on the same tasks: Anthropic reports Opus 5.5 at 66.4% on Terminal-Bench and 54.4% on FrontierCode, and different benchmarks are not directly comparable. As always, treat vendor numbers as a claim to be tested, not a fact. The pattern from earlier Grok releases is that real-world coding quality trails the published scores more than it does for Anthropic or OpenAI models.
Where Grok 4.7 makes sense
- High-volume, cost-sensitive workloads where a good-enough model at $6 beats a great one at $20
- Anything involving current events or X — Grok's real-time access to the platform is unique
- Long-document work that fits in 500K tokens without paying frontier prices
- Teams already paying for X Premium who get it bundled
Where it does not
- Regulated industries — xAI's safety documentation and enterprise controls remain thinner than its rivals'
- The hardest agentic coding, where Opus 5.5 and GPT-6 Astra lead
- Anyone uncomfortable with the model's willingness to be opinionated on contested topics
Grok 4.7 is the clearest example yet of the September pattern: the top of the market moving down in price faster than it moves up in capability. At $2/$6, it sets the floor every other frontier vendor now has to justify pricing above.
Related reading
Bottom line
Grok 4.7 is not the best model released in September, but it may be the best value, and for a large share of real workloads value is what matters. If you run high-volume text tasks, test it against your current model this week. If you need the last few points of coding reliability or enterprise assurances, the more expensive options still earn their price.
