Model Details
Grok-2
Quick answer: Grok-2 (released August 13, 2024) was xAI's frontier model at launch, introducing native image generation via the Aurora model integration, enterprise API access at $2/$10 per million tokens, and significantly improved reasoning over Grok-1.5. It was the first Grok model available for commercial API use at scale.
At a Glance
Where Grok-2 leads
- Native image generation via Aurora — text-to-image without a separate API call
- Real-time X (Twitter) knowledge, the defining xAI differentiator
- $2/$10 pricing — competitive with GPT-4o at launch
- Strong multilingual performance including French, German, Spanish, and Japanese
- Enterprise API with structured outputs, function calling, and system prompts
Where it lags
- 128K context (vs 200K+ for Claude and Gemini)
- SWE-bench Verified below the 2025–2026 frontier; superseded by Grok-3 and Grok-4
- Image generation via Aurora is a separate model — not natively integrated reasoning with generation
Best for: X-integrated applications, content generation with image output, and teams needing real-time social data alongside LLM capabilities.
What Grok-2 Is
Grok-2 was released at xAI's enterprise API launch in August 2024. It represented a major capability leap from Grok-1.5 — significantly better on GPQA (scientists-level questions), math, and coding, with the addition of vision understanding. The Aurora image generation integration made it the first commercial LLM API to offer combined text reasoning + image generation in a single provider relationship.
Grok-2 Mini launched simultaneously as a smaller, faster, cheaper variant for high-volume use cases where the full Grok-2 would be cost-prohibitive.
Specifications
| Field | Value |
|---|---|
| Organization | xAI |
| Context window | 128,000 tokens |
| License | Proprietary |
| Release date | August 13, 2024 |
| Knowledge cutoff | June 2024 |
| Modality | Multimodal (text + vision input; text + image output via Aurora) |
| Data access | Real-time X (Twitter) |
Pricing
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| xAI API | $2.00 | $10.00 |
Source: xAI API pricing.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| GPQA | 56.0% | xAI — Grok-2 | 2024-08 |
| MATH | 76.1% | xAI — Grok-2 | 2024-08 |
| HumanEval | 92.7% | xAI — Grok-2 | 2024-08 |
| MMLU | 87.5% | xAI — Grok-2 | 2024-08 |
Scores reported by xAI. Not Benchgen measurements.
Grok-2 vs Alternatives
| Model | Context | MATH | Price (in/out per 1M) |
|---|---|---|---|
| Grok-2 | 128K | 76.1% | $2 / $10 |
| Grok-2 mini | 128K | 70.1% | $0.20 / $1 |
| Grok-4 | 256K | — | $3 / $15 |
| GPT-4o (Aug 2024) | 128K | 76.6% | $2.50 / $10 |
(Rival scores from respective announcements; not Benchgen measurements.)
Related Models & Benchmarks
- Compare: Grok-2 mini · Grok-4 · GPT-4o
- Benchmarks: MMLU-Pro · SWE-bench Verified
Frequently Asked Questions
What is Grok-2?
Grok-2 is xAI's flagship model released August 2024. It introduced enterprise API access, native image generation via Aurora, real-time X data, and a major improvement in reasoning over Grok-1.5.How much does Grok-2 cost?
$2 per million input tokens and $10 per million output tokens via the xAI API.Specs sourced from xAI — Grok-2. Last updated 2026-06-19.
Grok-2
Grok-2 is a large language model developed by xAI.