Grok 4.7 enters API market behind Claude and GPT-6

Grok 4.7 arrives at $2 per million input tokens, giving xAI a lower-price answer to frontier AI models that still lead benchmark tables.

xAI introduced the model on September 21, 2026, describing it as its strongest system so far for coding and knowledge work. The company said Grok 4.7 uses a larger base model, longer reinforcement learning and improved self-checking of outputs.

Seven-point index gap

The pricing is the most immediate competitive feature. At $2 per million input tokens and $6 per million output tokens, Grok 4.7 is positioned below many Western frontier offerings and closer to lower-cost Chinese model pricing, according to the source material.

The performance picture is less favorable. On the Artificial Analysis Intelligence Index v4.3.2, which combines 10 benchmarks, Grok 4.7 scored 46, seven points behind Claude Fable 5.1 and GPT-6, which each scored 53.

That gap matters for developers weighing price against reliability. A lower token bill can reduce the cost of experiments and high-volume applications, but weaker benchmark results may limit adoption in workflows where accuracy, reasoning and tool use carry higher operational risk.

Coding test widens deficit

The largest shortfall appears in agentic coding. On Terminal-Bench 4.0, Grok 4.7 scored 26%, compared with 60% for GPT-6 Astra and 55% for Claude Fable 5.1.

The same test also placed DeepSeek V4.1 Flash at 27%, one percentage point ahead of Grok 4.7. That comparison is important because DeepSeek is also positioned as a lower-cost alternative, leaving xAI with price competition below and capability competition above.

Agentic coding tests are designed to assess whether a model can complete multi-step software tasks rather than only produce short code snippets. For buyers using AI tools inside engineering teams, that distinction can affect whether a model is used for drafts, debugging support or more autonomous coding work.

Distribution through developer tools

xAI is making Grok 4.7 available through the Grok API, Cursor and Grok Build. Those channels give the model a route into developer workflows where switching costs can be lower than in broader enterprise software deployments.

The distribution strategy also places Grok 4.7 in direct comparison with models already embedded in coding assistants and enterprise AI stacks. If developers can test the model cheaply in existing tools, xAI may gain usage data even where its benchmark position is not yet leading.

The company’s own framing emphasizes model scale, training duration and output verification. The independent benchmark results point to a different test: whether those technical changes translate into measurable gains against rival systems in tasks users can price and compare.

Low fees meet enterprise choices

For the AI sector, Grok 4.7 adds pressure to a market already split between premium performance and cheaper inference. If customers prioritize token costs for high-volume but lower-risk workloads, xAI’s pricing could pull more activity toward discounted frontier-adjacent models.

If benchmark results remain the main procurement filter, spending is more likely to stay concentrated around models with stronger scores in coding and reasoning tests. In that path, Grok 4.7 would need either better evaluations, narrower specialized use cases or further price cuts to shift enterprise decisions.

The macro effect is indirect but visible through business technology spending. Cheaper AI inference can lower software costs for companies adopting automation, while performance gaps can slow deployment in regulated, technical or customer-facing uses where errors are expensive.

The main uncertainty is whether Grok 4.7’s lower price is enough to overcome its measured deficit in agentic coding. If adoption rises through Cursor and the Grok API, xAI gains a larger developer base; if buyers stay with higher-scoring rivals, the release becomes another signal that AI competition is moving on both price and capability at once.