Grok is frontier-grade at coding, agents, and price — about 8/10 overall, with one caveat: verify its facts

Asked:How good is grok?

Evidence snapshot as of 1 September 2026: five authoritative pages from xAI, Vals AI, and Artificial Analysis, expanded into 125 extracted benchmark results for Grok 4.5 and 4.6. Independent testers place Grok 4.6 in line with the top frontier models on intelligence and near the very top on agentic work, at a markedly lower price — while flagging that benchmark strength does not guarantee factual reliability.

The scorecard, by use case

Qualitative synthesis of the evidence on a 0–10 scale — an editorial reading, not a normalized scientific score. Hover a row for the benchmarks behind it.

Grok 4.6 against its closest rivals

Score in % · xAI's own eval table, 12 Aug 2026 · higher is better · suites are not interchangeable · hover a dot for the exact value

Independent: Artificial Analysis scores Grok 4.6 at 61 on its Intelligence Index — in line with GPT-5.6 Sol — with a GDPval-AA v2 Elo of 1753, behind only Claude Opus 5, and 88.4% on Terminal-Bench v2.1. artificialanalysis.ai

Independent: Vals AI ranks Grok 4.5 #8 overall on the Vals Index (65.30%), but #4 on SWE-bench Verified (86.60%) and #5 on GPQA Diamond (92.93%) — stronger in code and science than in the average. vals.ai

Value is a genuine edge: on a Vals agent test Grok 4.5 was the cheapest model at $0.17 per test with 55 tasks resolved, and xAI prices it at $2 in / $6 out per million tokens at 80 tokens/sec — 4.6 pricing may differ. vals.ai · x.ai

The caveat: alongside Grok 4.5's accuracy gains, Artificial Analysis measured its hallucination rate rising from 25% to 54% — so consequential factual claims still need source checking. artificialanalysis.ai

Should you use it?

Yes, confidently

Coding and software agents (top-5 independent rankings), long-running agentic workflows, research that leans on live X data, and any workload where cost per task matters — Grok sits on the cost-performance frontier.

Yes, with checks

General reasoning and knowledge work: competitive rather than dominant. For unsupervised, high-stakes factual output, treat Grok as a strong drafter whose claims you verify — measured hallucination rates argue for a human in the loop.

The evidence rows

Model	Benchmark	Result	Tester	Kind	Source
Grok 4.6	AA Intelligence Index	61	Artificial Analysis	independent	artificialanalysis.ai
Grok 4.6	GDPval-AA v2 Elo	1753 — behind only Claude Opus 5	Artificial Analysis	independent	artificialanalysis.ai
Grok 4.6	Terminal-Bench v2.1	88.4%	Artificial Analysis	independent	artificialanalysis.ai
Grok 4.6	𝜏³-Banking	50.7% — top two with Qwen3.8 Max	Artificial Analysis	independent	artificialanalysis.ai
Grok 4.6	AA-Briefcase Elo	1577 (debut)	Artificial Analysis	independent	artificialanalysis.ai
Grok 4.5	Vals Index	65.30% — #8 overall	Vals AI	independent	vals.ai
Grok 4.5	SWE-bench Verified	86.60% — #4	Vals AI	independent	vals.ai
Grok 4.5	GPQA Diamond	92.93% — #5	Vals AI	independent	vals.ai
Grok 4.5	Vibe Code Bench	69.00% — #10 (up from #47)	Vals AI	independent	vals.ai
Grok 4.5	Harvey Legal Agent Bench	12.92% — #2	Vals AI	independent	vals.ai
Grok 4.5	Cost / task (Vals agent test)	$0.17 — cheapest, 55 tasks resolved	Vals AI	independent	vals.ai
Grok 4.6	CursorBench v3.2	69.9%	xAI	vendor	x.ai
Grok 4.6	DeepSWE v1.1	65.9%	xAI	vendor	x.ai
Grok 4.6	APEX-Agents	57.5%	xAI	vendor	x.ai
Grok 4.5	API price & speed	$2 in / $6 out per 1M tokens · 80 TPS	xAI	vendor	x.ai

Method: synthesis of 5 authoritative source pages (xAI, Vals AI, Artificial Analysis) expanded into 125 extracted model/benchmark rows, snapshot as of 2026-09-01. Independent rows (Vals AI, Artificial Analysis) are prioritized; vendor rows are xAI's own eval table. Benchmark suites use different tasks and scales and are not directly comparable; Elo and index scores are unitless. Duplicate and weakly labeled extractions were cut for space. The 0–10 scorecard is a qualitative editorial synthesis, not a measured quantity. Grok 4.5 pricing does not necessarily apply to Grok 4.6.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.