Grok 4.1
- Provider
- xAI
- Status
- Superseded
- Context
- 256,000 tok
Grok 4.1 is the xAI model released in November 2025 (after a quiet 1–14 November rollout) that vaulted Grok to the top of the public arena. Its Thinking mode took #1 on LMArena at 1483 Elo, with Non-Thinking #2 at 1465 — a roughly 31-point lead over third place and a jump from Grok 4’s ~#33 (~1409 Elo), one of the largest ranking gains in the leaderboard’s history (Elon Musk on X, NextBigFuture).
Rather than a new reasoning base, Grok 4.1 was a refinement of Grok 4 focused on conversational quality, emotional intelligence and reliability: xAI reported a ~65% reduction in hallucination (from about 12.1% to 4.2%), and in blind tests users preferred its responses about 64.8% of the time (Fello AI, Codecademy). It has since been superseded by Grok 4.2 and Grok 4.3.
Quick specs
| Provider | xAI |
| Released | November 2025 |
| Status | Superseded (available via API) |
| API model ID | grok-4.1 |
| Context window | 256,000 tokens (2M on Grok 4.1 Fast) |
| LMArena Elo | 1483 (#1, Thinking) / 1465 (#2, Non-Thinking) |
| Hallucination | ~4.2% (down ~65% from Grok 4’s ~12.1%) |
| Best for | Conversational quality, everyday use, real-time X data (historical) |
| Limitations | Superseded by Grok 4.2/4.3; reasoning base carried over from Grok 4 |
What changed in Grok 4.1
Grok 4.1 kept Grok 4’s technical reasoning capability and improved the things users actually feel day to day (Codecademy):
- Much lower hallucination — a roughly 65% reduction, from about 12.1% to 4.2%, addressing Grok’s biggest reliability complaint.
- Stronger emotional intelligence and creative writing — warmer, more natural responses, which is what drove its arena win.
- Thinking and Non-Thinking modes — both ranked at the very top of LMArena.
- A 2M-context Grok 4.1 Fast variant for high-volume and long-document use.
The result was the highest-rated model on LMArena at the time, even though its underlying reasoning benchmarks were broadly inherited from Grok 4 — a reminder that arena preference rewards conversational quality, not just raw capability.
Benchmark performance
| Benchmark | Grok 4.1 | Notes |
|---|---|---|
| LMArena Elo (Thinking) | 1483 (#1) | ~31-point lead over third place |
| LMArena Elo (Non-Thinking) | 1465 (#2) | Both modes at the top |
| Hallucination rate | ~4.2% | Down ~65% from Grok 4 (~12.1%) |
The story is the leaderboard jump and the reliability gain, not a new reasoning ceiling (NextBigFuture). Arena Elo drifts over time and later models (including xAI’s own Grok 4.2 and Grok 4.3) moved ahead, so treat the #1 ranking as a November-2025 snapshot. See best AI models for current standings.
Pricing and access
Grok 4.1 was available free with daily limits in the Grok app, and with higher limits via SuperGrok ($30/month) and X Premium+ ($40/month) (Codecademy). Via the API it followed Grok 4’s pricing tier, with a cheaper Grok 4.1 Fast variant (2M context) for high-volume use.
How Grok 4.1 compares
- vs Grok 4 — same reasoning base, but far better conversational quality, ~65% less hallucination, and a leap from ~#33 to #1 on LMArena.
- vs its successors — Grok 4.2 added stronger coding, and Grok 4.3 added native video input and is the current flagship; prefer them for new work.
- vs 2025 rivals — At release it topped LMArena over Gemini 2.5 Pro and the leading OpenAI and Anthropic models on human-preference voting, though that is preference-based, not a capability sweep.
Known limitations
Superseded. Grok 4.1 is behind xAI’s current Grok 4.3. Reasoning carried over. Its gains were in conversational quality and reliability, not a new reasoning ceiling. Arena rankings are snapshots. The #1 position reflects November 2025 and has since been overtaken. The wider Grok safety and policy context applies — see the xAI provider page.
FAQ
What is Grok 4.1?
Grok 4.1 is xAI’s November 2025 model — a refinement of Grok 4 that topped LMArena (1483 Elo, #1) and cut hallucinations by about 65%. It improved conversational quality and emotional intelligence rather than introducing a new reasoning base, and has since been superseded by Grok 4.2 and 4.3.
Why was Grok 4.1 ranked #1 on LMArena?
LMArena ranks by human preference, and Grok 4.1’s warmer, more reliable, less hallucination-prone responses won blind comparisons about 64.8% of the time. Its Thinking mode reached 1483 Elo, a ~31-point lead over third place.
How much did Grok 4.1 cost?
It was free with daily limits in the Grok app, with higher limits via SuperGrok ($30/month) and X Premium+ ($40/month). The API followed Grok 4’s pricing, with a cheaper 2M-context Grok 4.1 Fast variant.
Is Grok 4.1 still current?
No — it has been superseded by Grok 4.2 and Grok 4.3. It remains available via the API.
Last verified 18 June 2026. Grok 4.1’s figures are xAI-reported (via Elon Musk’s announcement, NextBigFuture, Fello AI and Codecademy); LMArena rankings are human-preference snapshots that change over time. Confirm against xAI’s official pages before relying on specific numbers.