xAI has released Grok 4.7, its most capable reasoning model, built on a new, larger base model trained with extended reinforcement learning on multi-hour difficulty tasks. The headline pitch: frontier performance at commodity pricing. At $2 per million input tokens and $6 per million output tokens, it undercuts GPT-5.6 Sol by 2-3x and Fable 5.1 by 5-8x on token cost. The model is available immediately in Cursor, Grok Build, and through the Grok API. The benchmark story is genuinely mixed, and xAI deserves some credit for publishing numbers that don't uniformly flatter. Grok 4.7 scores 46.3% on CursorBench 4.0, beating GPT-5.6 Sol (41.7%) and its own predecessor (40.4%) but trailing Fable 5.1's 51.8%. On DeepSWE v1.1, the asterisked 71.0% "high effort" score lands between GPT-5.6 Sol (72.7%) and Fable 5.1 (70.0%) — though that asterisk matters and the comparison is not apples-to-apples. The standout results are in electrical engineering (EEBench 64.0%, well ahead of all competitors) and legal work (Harvey Legal Agent Benchmark 19.6% vs. GPT-5.6 Sol's dismal 2.5%). Where Grok 4.7 clearly lags is on multi-hour terminal work (Terminal-Bench 4.0: 38.0% vs. Fable 5.1's 57.9%) and clinical reasoning (HealthBench Professional: 56.7% vs. Fable's 62.1%). The model does not uniformly dominate any competitor across all benchmarks — it trades punches, winning some rounds and losing others. The safety narrative is more assertive. xAI claims Grok 4.7 has an "entirely new safeguard stack" and leads on jailbreak resistance, with only 3.3% of risky dual-use prompts passing through on HackerBench v0.3. The LatchBio biosafety score of 62.4% is cited as best-in-class. These are self-reported numbers from proprietary benchmarks in several cases, so the integrity of the claims depends on trust in xAI's internal testing regime. The pricing strategy is the clearest signal of xAI's competitive positioning. By holding Grok 4.7 at identical pricing to Grok 4.6 while delivering measurable improvements, xAI is competing on value rather than pure capability. This is a deliberate move to capture the cost-sensitive developer segment — teams building coding agents in Cursor, API-driven workflows, and enterprise knowledge work — where Fable 5.1's $10/$50 pricing creates real friction. The broader picture: the frontier model market is consolidating around a pattern where no single model wins everywhere, and price-performance becomes the real differentiator. Grok 4.7 slots into this dynamic as a strong cost-leader with genuine spikes in legal and EE domains, but it does not displace Fable 5.1 or GPT-5.6 Sol as the overall capability leader. For developers building on API pricing, the 2-5x cost advantage over competitors is the actual story.
