Grok 4.2
- Provider
- xAI
- Status
- Superseded
- Context
- 256,000 tok
- SWE-bench
- 75%
- Knowledge
- 2025-11
Grok 4.2 is xAI’s latest model, launched on 17 February 2026 as a public beta and release candidate. xAI positions it as a frontier-class reasoning model with a headline focus on coding and agentic performance: at launch the company reports roughly 75% on SWE-bench, ahead of its cited figures for GPT-5 (74.9%) and Claude Opus 4 (74.5%), and about 15.9% on ARC-AGI, which xAI describes as the first model to break the 10% mark on that abstract-reasoning benchmark (computertech).
Grok 4.2 is reported to use roughly a trillion parameters with a 256K-token context window, and it carries forward Grok’s signature edge — real-time access to X data via DeepSearch. As a release candidate, it is available first through Grok’s consumer subscriptions rather than a standalone per-token API price. The figures below are xAI’s own launch claims and had not been independently verified at release.
Quick specs
| Provider | xAI |
| Released | 17 February 2026 (public beta / release candidate) |
| Context window | 256,000 tokens |
| Parameters | ~1 trillion (reported) |
| Knowledge cutoff | November 2025 |
| Modalities | Text and images in; text out |
| SWE-bench | ~75% (xAI launch claim) |
| ARC-AGI | ~15.9% (xAI launch claim) |
| Access | SuperGrok ($30/mo), X Premium+ ($40/mo) |
| Best for | Coding and agentic tasks, reasoning, real-time X research |
| Limitations | Vendor-reported benchmarks only; beta/RC at launch; no standalone API price yet |
What’s new in Grok 4.2
Grok 4.2 is an iteration on the Grok 4 line, sharpening coding and agentic performance rather than overhauling the architecture (computertech).
- Coding focus. xAI’s headline claim is SWE-bench leadership at launch — roughly 75%, narrowly ahead of its cited GPT-5 (74.9%) and Claude Opus 4 (74.5%) figures.
- Abstract reasoning. xAI reports about 15.9% on ARC-AGI and frames Grok 4.2 as the first model to clear 10% on that benchmark.
- Agentic behaviour. In the Alpha Arena trading simulation, xAI reports Grok 4.2 returning 12%-plus while some rival models lost money — an agentic-decision showcase rather than a standardised benchmark.
- Real-time data. Grok 4.2 keeps DeepSearch and live access to X, xAI’s structural advantage over web-search-only rivals.
Benchmark performance
All figures are xAI’s own launch claims (February 2026) and were not independently verified at release. Where a number is not provided, it is left as data not available rather than estimated.
| Benchmark | Grok 4.2 (xAI) | xAI’s cited comparison |
|---|---|---|
| SWE-bench | ~75% | GPT-5 74.9%, Claude Opus 4 74.5% |
| ARC-AGI | ~15.9% | “First model to break 10%“ |
| Alpha Arena | 12%+ return | Some rivals reportedly lost money |
The launch framing is coding-first, and the numbers are close enough to GPT-5 and Claude Opus 4 that the honest read is parity-to-slight-lead on xAI’s chosen harness rather than a decisive jump — and vendor launch numbers typically sit above standardised leaderboards. Independent SWE-bench Pro and Artificial Analysis figures for Grok 4.2 were not available at launch, so a cross-model consensus placement should wait for third-party testing.
How to access Grok 4.2
At launch, Grok 4.2 is a public beta / release candidate available through Grok’s consumer subscriptions; xAI has not published a standalone per-token API price for it.
| Tier | Price | Grok 4.2 | Notes |
|---|---|---|---|
| Free | $0 | No | Earlier Grok models only |
| SuperGrok | $30/mo ($300/yr) | Yes | Standalone app; DeepSearch, voice |
| X Premium+ | $40/mo | Yes | Grok inside X, bundled with X features |
| SuperGrok Heavy | $300/mo | Yes | Adds the multi-agent Grok 4 Heavy |
Grok is also available inside the X app and in Tesla vehicles. See the Grok app page for the full consumer breakdown and the xAI provider page for company context.
How Grok 4.2 compares
At launch, Grok 4.2’s frontier contemporaries are Anthropic’s Claude Opus 4.6, OpenAI’s GPT-5 line and Google’s Gemini 3 Pro. xAI’s own comparison set cites GPT-5 (74.9% SWE-bench) and Claude Opus 4 (74.5%), with Grok 4.2 narrowly ahead at ~75% on its harness. Because these are vendor figures on xAI’s chosen benchmark, the safest read is that Grok 4.2 is competitive with the current frontier on coding rather than clearly ahead of it; independent, same-harness numbers are not yet available. Grok’s durable differentiator remains real-time X data, not raw benchmark leadership. See best AI models and best AI for coding for where it sits once independent testing lands.
Known limitations
Vendor-reported benchmarks only. Every headline figure is an xAI launch claim on its own harness; independent SWE-bench Pro and Artificial Analysis results were not available at release.
Release candidate. Grok 4.2 launched as a public beta / RC, so behaviour and limits may change before a stable release.
No standalone API price at launch. Access is via SuperGrok and X Premium+ subscriptions; per-token API pricing is data not available.
Same safety and policy context as the wider Grok line. xAI publishes less safety documentation than its peers, and the Grok apps’ image tools have been the subject of regulatory scrutiny in 2026 — see the xAI provider page.
Frequently asked questions
When was Grok 4.2 released?
xAI launched Grok 4.2 on 17 February 2026 as a public beta and release candidate, available through SuperGrok and X Premium+.
How good is Grok 4.2 at coding?
At launch, xAI reported roughly 75% on SWE-bench — narrowly ahead of its cited figures for GPT-5 (74.9%) and Claude Opus 4 (74.5%). These are vendor numbers on xAI’s own harness and were not independently verified at release, so treat them as a ceiling.
What is Grok 4.2’s context window?
A 256,000-token context window, with text and image input and text output. It is reported to use roughly a trillion parameters, though xAI has not officially confirmed the figure.
How do I access Grok 4.2?
Through Grok’s consumer subscriptions — SuperGrok ($30/month or $300/year) and X Premium+ ($40/month) — plus the X app and Tesla vehicles. xAI did not publish a standalone per-token API price at launch.
Is Grok 4.2 better than Claude or GPT-5?
On xAI’s launch benchmarks it is roughly level with GPT-5 and Claude Opus 4 on SWE-bench, with a slight reported lead. Independent, same-harness comparisons were not available at launch, so a definitive consensus placement should wait for third-party testing. Grok’s clearest advantage remains real-time X data.
Written at launch and last verified 17 February 2026. All benchmark figures are xAI’s own launch claims and were not independently verified at release; independent figures are marked data not available. Pricing and availability reflect the February 2026 launch and are subject to change.