xAI Releases Grok 4.7 With a Bigger Model and the Same API Pricing

xAI has released Grok 4.7, updating its flagship model for coding, professional knowledge work and other demanding tasks while keeping API pricing unchanged from Grok 4.6.

xAI announced the model on September 21, saying Grok 4.7 uses a larger base model and longer reinforcement-learning training. The company lists API pricing at $2 per million input tokens and $6 per million output tokens, preserving the pricing of the previous generation.

The release is notable because frontier-model competition is increasingly about more than raw capability. Developers now compare coding performance, latency, context handling, tool use and the cost of serving large numbers of requests.

What changed in Grok 4.7

xAI says the new model builds on Grok 4.6 with a larger foundation and additional reinforcement-learning training. Longer post-training can improve how a model follows instructions, reasons through complex tasks and uses tools.

The company is positioning Grok 4.7 particularly strongly for coding and professional work, two areas where users are often willing to pay for improvements because better model output can save meaningful labor time.

Independent evaluation will be important. Vendor benchmarks can provide useful information, but real-world performance depends heavily on the exact workflow.

Keeping API pricing flat matters

Model capability tends to improve rapidly, but businesses care about cost per useful task rather than benchmark scores alone. If a new model completes work more reliably at the same token price, its effective economics improve.

That is particularly important for AI agents. An agent may make many model calls to complete one job, so small differences in token cost can compound across large deployments.

At the same time, token pricing is only one component of cost. A model that uses more output tokens or requires repeated attempts can be more expensive even if its published rate is lower.

Coding remains one of the most competitive AI markets

Software development has become a major proving ground for frontier models because results can often be tested objectively. Code either compiles and passes tests or it does not.

Developers increasingly use AI for code generation, debugging, documentation, refactoring and navigating unfamiliar repositories. More advanced systems can operate as agents that modify files, run tests and iterate on errors.

That creates a large commercial opportunity because software engineers are expensive and small productivity gains can justify substantial AI spending.

Professional knowledge work is harder to evaluate

Tasks such as research, analysis and document preparation are less deterministic than coding. A response can look polished while containing subtle errors.

For professional use, factual reliability and source handling therefore matter as much as writing quality. Organizations need review processes that match the consequences of an error.

A model used to brainstorm marketing copy can tolerate more uncertainty than one used to summarize a contract or support a financial decision.

xAI is competing in a crowded frontier market

Grok competes with models from OpenAI, Anthropic, Google, Meta and a growing number of Chinese developers. That competition is pushing vendors to release new generations quickly while cutting or holding prices.

The result benefits developers but makes platform selection more complicated. The “best” model can vary by task, and performance differences may change with each release.

Companies increasingly route different workloads to different models rather than relying exclusively on one provider.

Model size is not enough to determine quality

xAI’s description of a larger base model signals additional capacity, but parameter count by itself does not establish that one system is better than another. Training data, architecture, reinforcement learning and inference techniques all influence performance.

Efficiency matters too. A smaller model that solves a task with fewer tokens and lower latency may be more useful in production than a larger model with slightly better benchmark results.

What developers should test

Teams considering Grok 4.7 should evaluate it against their own workloads. For coding, that can include repository-level tasks, test generation and bug fixing. For research, teams can measure citation accuracy, instruction following and consistency.

Latency and rate limits matter in production, especially for customer-facing applications. Developers should also calculate the full cost per completed task rather than comparing token rates in isolation.

Model switching should be designed into applications where practical. The rapid pace of releases means today’s preferred model may not remain the best option for long.

Bottom line

Grok 4.7 shows how quickly the frontier-model market is evolving. xAI is offering a larger, more heavily trained model while holding API pricing at $2 per million input tokens and $6 per million output tokens.

The important question is whether those improvements translate into better economics for real work. Coding and agentic workflows provide relatively measurable tests. Professional knowledge work requires more careful verification.

As competition intensifies, developers are likely to benefit from stronger models at lower effective cost. The winners will not necessarily be the systems with the biggest headline specifications, but the ones that complete useful tasks reliably enough and cheaply enough to remain in production.