Grok 4
- Provider
- xAI
- Status
- Superseded
- Context
- 256,000 tok
- Price
- $3 / $15 /MTok
Grok 4 is the xAI model released on 9 July 2025 that established Grok as a genuine frontier contender. It was xAI’s first always-on reasoning model — it “thinks” before every response — and it launched alongside a multi-agent variant, Grok 4 Heavy. Trained on xAI’s Colossus supercomputer, Grok 4 posted frontier-class reasoning and maths scores: independent evaluator Artificial Analysis gave it an Intelligence Index of 73 at release, then ahead of OpenAI’s o3, Google’s Gemini 2.5 Pro and Anthropic’s Claude 4 Opus.
Grok 4 has since been superseded — by Grok 4.1 (November 2025), then Grok 4.2 and Grok 4.3 — but it remains available via the API and is the model that put xAI on the frontier map. It carries a 256,000-token context window and priced at $3/$15 per million tokens at launch (llm-stats).
Quick specs
| Provider | xAI |
| Released | 9 July 2025 |
| Status | Superseded (available via API) |
| API model ID | grok-4 |
| Context window | 256,000 tokens |
| Input price | $3.00 / MTok |
| Output price | $15.00 / MTok |
| GPQA Diamond | ~87% |
| Humanity’s Last Exam | 25.4% (no tools) / 44.4% (with tools) |
| ARC-AGI-2 | 15.9% |
| AA Intelligence Index | 73 (at release) |
| Best for | Maths and reasoning, real-time X data (historical) |
| Limitations | Superseded by Grok 4.1/4.2/4.3; vendor-run headline benchmarks |
What Grok 4 was
Grok 4 was the release where xAI moved from “fast chatbot” to frontier reasoning lab. Built on the Colossus cluster, it reasons before answering and integrates xAI’s signature real-time access to X data. DataCamp’s hands-on coverage and xAI’s own announcement framed it as a step change over Grok 3, and the independent Artificial Analysis Intelligence Index of 73 backed that up at launch.
It shipped with a sibling, Grok 4 Heavy, a multi-agent “study group” configuration for the hardest problems, available on the $300/month SuperGrok Heavy tier.
Benchmark performance
xAI-reported launch figures with independent context from Artificial Analysis; treat vendor numbers as a ceiling.
| Benchmark | Grok 4 | Notes |
|---|---|---|
| GPQA Diamond | ~87% | Graduate-level science |
| Humanity’s Last Exam | 25.4% / 44.4% | No tools / with tools |
| ARC-AGI-2 | 15.9% | ~Double the prior best at release |
| AA Intelligence Index | 73 | Ahead of o3, Gemini 2.5 Pro, Claude 4 Opus at release |
Grok 4’s strengths were maths and reasoning; the multi-agent Grok 4 Heavy pushed these further (100% on AIME 2025, 50.7% on Humanity’s Last Exam). On LMArena, Grok 4 initially sat around 1409 Elo before Grok 4.1 vaulted the line to #1 in November 2025 (NextBigFuture).
Pricing
| Input (per MTok) | Output (per MTok) | |
|---|---|---|
| Grok 4 | $3.00 | $15.00 |
At launch Grok 4 cost $3/$15 via the xAI API. Consumer access came through SuperGrok ($30/month), with Grok 4 Heavy on SuperGrok Heavy ($300/month) (llm-stats). A cheaper Grok 4 Fast variant followed for high-volume use.
How Grok 4 compares
- vs its successors — Grok 4.1 sharply improved conversational quality and cut hallucinations and topped LMArena; Grok 4.2 and Grok 4.3 pushed coding and added native video input. For current use, prefer the newer models.
- vs 2025 rivals — At release Grok 4 led the Artificial Analysis Intelligence Index over o3, Gemini 2.5 Pro and Claude 4 Opus, a position later overtaken as the field advanced.
Known limitations
Superseded. Grok 4 is two-plus generations behind xAI’s current Grok 4.3; use it only for continuity. Vendor-run headline benchmarks. The maths/reasoning launch figures are xAI’s own. Same safety context as the wider Grok line — see the xAI provider page for xAI’s thinner safety documentation and the 2026 image-generation controversy.
Version history
| Version | Released | Key points |
|---|---|---|
| Grok 4 / Grok 4 Heavy | 9 Jul 2025 | Always-on reasoning; AA Intelligence Index 73; multi-agent Heavy |
| Grok 4.1 | Nov 2025 | Topped LMArena; big hallucination cut |
| Grok 4.2 | Feb 2026 | Stronger coding |
| Grok 4.3 | Apr 2026 | Current flagship; native video input |
FAQ
What is Grok 4?
Grok 4 is xAI’s July 2025 always-on reasoning model — the release that made Grok a frontier contender, with an Artificial Analysis Intelligence Index of 73 at launch. It has since been superseded by Grok 4.1, 4.2 and 4.3.
How much did Grok 4 cost?
$3 per million input tokens and $15 per million output via the xAI API, with consumer access through SuperGrok ($30/month) and Grok 4 Heavy on SuperGrok Heavy ($300/month).
What’s the difference between Grok 4 and Grok 4 Heavy?
Grok 4 Heavy is a multi-agent “study group” configuration that runs several agents in parallel for the hardest reasoning, scoring higher than standard Grok 4 (for example, 100% on AIME 2025). It was limited to the SuperGrok Heavy tier.
Is Grok 4 still worth using?
For new work, no — xAI’s current Grok 4.3 is far ahead. Grok 4 remains available via the API mainly for continuity.
Last verified 18 June 2026. Grok 4’s headline figures are xAI-reported from the July 2025 launch, with independent context from Artificial Analysis and llm-stats. Pricing and standing change quickly — confirm against xAI’s official pages before relying on specific numbers.