Grok 4.7 ships at $2/$6 with a 2.1T base and lands mid-pack on the Intelligence Index

SpaceXAI shipped Grok 4.7 on September 21, 2026, at $2 per million input tokens and $6 per million output, the same prices Grok 4.6 carried, on a base model that has grown 40%, from 1.5 trillion parameters to 2.1 trillion. It's a bigger model at a held price, and that framing is doing most of the marketing work.

The independent scoreboards tell the other half. Artificial Analysis Intelligence Index v4.3.2, a composite of 10 benchmarks, puts Grok 4.7 at 46. Claude Fable 5.1 and GPT-6 Astra both score 53. On Terminal-Bench 4.0, per Artificial Analysis via The Decoder, Grok 4.7 lands at 26%, against roughly 60% for GPT-6 Astra and 55% for Fable 5.1. DeepSeek V4.1 Flash sits at 27%, a hair above Grok on the same test.

SpaceXAI's own model card reports 38.0% on Terminal-Bench 4.0 at xhigh effort under its Grok Build harness, and OfficeChai's coverage cites 71.0% on DeepSWE v1.1 at high effort, plus xHigh results of 64.0% on EEBench, 56.7% on HealthBench Professional, and 19.6% on the Harvey Legal Agent Benchmark. Fable 5.1, for reference, scores 51.8% on CursorBench 4.0 and 57.9% on Terminal-Bench 4.0 in SpaceXAI's own comparison table. Read together, Grok 4.7 wins on price and loses on the reasoning-heavy agentic evals that decide production suitability.

The pricing context is the actual news. Fable 5.1 and GPT-6 Astra both charge $50 per million output tokens. Grok's $6 undercuts them by more than 8x. Google is holding Gemini 3.8 Flash at $0.75 input and $3.75 output through the end of 2026. Decrypt's headline calls Grok "Late to the AI Frontier Party." The Neuron calls it a price war. Both are right.

For a small-business owner, none of this is actionable on its own. Every release adds another matrix of harnesses, reasoning-effort dials, benchmark versions, and per-token tradeoffs. A founder pricing out CursorBench 4.0 differentials at xHigh isn't finding customers that hour.

That's the opening LemonLime is built for. It's a done-for-you service that decides which customer-growth work matters, prepares it, and delivers finished output at 9:00 AM local time. The owner doesn't pick the model or track the price war; the frontier's advances flow through without a spreadsheet.

The pattern is familiar from prior weeks: DeepSeek's V4.1 Flash architecture reset, Anthropic's Fable 5.1 automation jump, and GPT-6 Astra's computer-use launch each rewrote a different corner of the stack. Grok 4.7 rewrites price. The frontier isn't consolidating. It's fragmenting, on a monthly clock.

Sources

• https://media.x.ai/v1/website/4p7card-5eccc980.pdf
• https://the-decoder.com/xai-launches-grok-4-7-at-bargain-prices-but-benchmarks-reveal-a-wide-gap-to-claude-and-gpt-6/
• https://officechai.com/ai/grok-4-7-benchmarks/
• https://decrypt.co/378824/xai-launches-grok-4-7
• https://www.theneuron.ai/news/xais-grok-47-makes-frontier-ai-a-price-war/