xAI Unveils Grok 4.3

xAI has unveiled Grok 4.3. It scored 53 points on the Artificial Analysis Intelligence Index, surpassing Muse Spark and Claude Sonnet 4.6, and 4 points higher than its immediate predecessor, Grok 4.20 0309 v2.

Cost efficiency also improved. The cost to run the full benchmark suite is $395, about 20% lower than Grok 4.20 0309 v2. This reflects a 37.5% cut in input token pricing and a 58.3% cut in output token pricing; although it uses about 44% more output tokens, it remains relatively cheap for its intelligence tier.

In real-world agentic performance, GDPval-AA saw the biggest jump.

• Its ELO is 1500, up 321 points from the previous version's 1179.
• It surpassed Gemini 3.1 Pro Preview, Muse Spark, GPT-5.4 mini (xhigh), and Kimi K2.5, but trails GPT-5.5 (xhigh) by 276 points.
• The estimated win rate on a standard Elo basis is about 17%.

In other evaluations, 𝜏²-Bench Telecom rose 5 points to 98%, while IFBench held steady at 81%. On the other hand, AA-Omniscience Accuracy rose 8 points but the Non-Hallucination Rate fell 8 points, showing the balance between knowledge accuracy and hallucination suppression is still not perfect. As a result, Grok 4.3 has moved onto the intelligence-versus-cost Pareto frontier.