Overview
Grok 1.5 is the second model in xAI's flagship Grok line, announced on March 28, 2024 and rolled out to Grok users on the X (formerly Twitter) platform in the weeks that followed. It was positioned as a reasoning-focused upgrade to Grok 1, the model xAI later open-sourced under Apache 2.0. Unlike Grok 1, the weights for Grok 1.5 were never released — it is a proprietary model.
The headline change in Grok 1.5 was its context window, which jumped to 128,000 tokens — roughly a sixteen-fold increase over Grok 1. In xAI's Needle-in-a-Haystack retrieval tests, Grok 1.5 recovered embedded information with perfect accuracy across the full 128K-token window. The model also posted clear gains on math and coding evaluations, with xAI reporting 50.6% on MATH, 90% on GSM8K and 74.1% on HumanEval.
Grok 1.5 reached users only through the Grok chatbot for X Premium subscribers; xAI did not publish a developer API or per-token pricing for the text model at this stage (the public xAI API arrived later with the Grok 2 generation). Grok 1.5 was superseded by Grok 2 in August 2024 and is now of mainly historical interest as the version where xAI's flagship line first reached long-context, near-GPT-4-tier reasoning.
| Released | 2024-05-15 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Not disclosed |
| Context | 128K tokens |
| Max output | Not published by xAI |
| Architecture | Proprietary transformer-based large language model, an iteration on Grok 1. xAI trained it on a custom distributed training stack built on JAX, Rust and Kubernetes. The exact parameter count and whether 1.5 retained Grok 1's mixture-of-experts design were not disclosed. |
| Knowledge cutoff | Not published by xAI |
| Modalities | Text |
| Status | Superseded — replaced by Grok 2 (August 2024) and later flagship Grok releases. The text model was never offered through a public API and is no longer the current Grok. |
Benchmarks
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Strengths
- 128K-token context window — a 16x jump over Grok 1 — with perfect Needle-in-a-Haystack retrieval across the full window
- Strong grade-school-to-competition math: 90% on GSM8K and 50.6% on the harder MATH set
- Solid code generation for its era (74.1% on HumanEval)
- Marked reasoning improvement over Grok 1, bringing xAI's flagship line close to GPT-4-tier performance
Best for
- Long-document analysis and retrieval that needed a large context window in 2024
- Math and reasoning tasks, from word problems to competition-style problems
- Code generation and programming help inside the Grok chatbot
- General conversational assistance for X Premium subscribers
Grok (flagship) — every version
The full lineage of the Grok (flagship) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
FAQ
When was Grok 1.5 released?
xAI announced Grok 1.5 on March 28, 2024, and made it available to Grok users on the X platform over the following weeks (around May 2024).
What is Grok 1.5's context window?
Grok 1.5 has a 128,000-token context window — about sixteen times larger than Grok 1's. In xAI's Needle-in-a-Haystack tests it retrieved embedded information with perfect accuracy across the full window.
Are Grok 1.5's weights open?
No. Unlike Grok 1, whose weights xAI open-sourced under Apache 2.0 in March 2024, Grok 1.5 is a proprietary model and its weights were never released.
Could you use Grok 1.5 through an API?
No public API or per-token pricing was published for the Grok 1.5 text model. It was accessible only through the Grok chatbot for X Premium subscribers; xAI's developer API came later with the Grok 2 generation.