Grok 4.3

Summary

Grok 4.3 is xAI's frontier reasoning model, released May 1, 2026, ranking #1 on CaseLaw v2 (79.3%) and CorpFin benchmarks while scoring 98% on τ²-Bench Telecom. At $1.25/$2.50 per million tokens, it undercuts Claude Sonnet 4.6's input cost by roughly 5x.

Overview

Grok 4.3 is xAI's frontier reasoning model, released May 1, 2026 (general availability) following an April 17, 2026 beta. It marks a major step up from Grok 4.20 — adding stronger agentic performance, dramatically lower pricing, and a leap in domain benchmarks (notably legal, finance, and telecom). On the Artificial Analysis Intelligence Index it scores 53, four points ahead of Grok 4.20 0309 v2 and just above [[Anthropic/Claude Sonnet 4.6|Claude Sonnet 4.6]] and [[Meta AI/Muse Spark|Muse Spark]].

The release pairs an aggressively low API price with a new fast/powerful voice cloning suite, signalling xAI's strategy of competing on price-performance rather than raw frontier rank while [[xAI/Grok 5|Grok 5]] (6T parameter MoE) continues training on the Colossus 2 supercluster.

Specifications

  • Developer: xAI
  • Model String: grok-4.3 / grok-4.3-beta
  • Release Date: April 17, 2026 (Beta); May 1, 2026 (GA)
  • Type: Reasoning LLM, multimodal (text + vision)
  • Pricing: $1.25 per 1M input tokens / $2.50 per 1M output tokens (~20% lower than Grok 4.20 0309 v2)
  • Cost on AAI Index: ~$395 to run the full Artificial Analysis Intelligence Index
  • Access: xAI API, OpenRouter, Oracle OCI Enterprise AI, Grok app + x.com

Capabilities

Agentic Reasoning Step-Up: Largest single benchmark improvement is on GDPval-AA, where Grok 4.3 scores ELO 1500 — up 321 points from Grok 4.20 0309 v2 (1179) — surpassing Gemini 3.1 Pro Preview, Muse Spark, GPT-5.4 mini (xhigh), and Kimi K2.5.

Domain Leadership: #1 on CaseLaw v2 (79.3% accuracy) and #1 on CorpFin — making Grok 4.3 the strongest available model for legal and corporate-finance reasoning per Artificial Analysis tracking.

Tool Use & Telecom: Scores 98% on τ²-Bench Telecom and 81% on IFBench — among the best published results for instruction-following and tool-using agents.

Voice Cloning Suite: Released alongside Grok 4.3 — fast, low-latency, high-fidelity multi-speaker voice cloning, deepening xAI's multimodal stack.

Pricing-Performance Frontier: At $1.25 / $2.50 per 1M tokens, Grok 4.3 is among the cheapest frontier-tier models, undercutting Claude Sonnet 4.6 ($3 / $15) by ~5x on input cost.

Limitations

Grok 4.3 sits below the absolute reasoning frontier on hardest-class benchmarks ([[Google DeepMind/Gemini 3.1 Ultra|Gemini 3.1 Ultra]], [[Anthropic/Claude Opus 4.7|Claude Opus 4.7]], [[OpenAI/GPT-5.5|GPT-5.5]]). The competitive moat is more on price-performance than on raw capability ceiling, and the model still inherits Grok's broader brand-and-policy questions around content moderation choices.

Recent Developments

  • April 17, 2026 — Grok 4.3 beta released; immediately available via OpenRouter, Kilo Code, and Puter Developer.
  • May 1, 2026 — General availability; new pricing and voice cloning suite shipped alongside.
  • May 2026 — Grok 4.3 added to Oracle Cloud Infrastructure Enterprise AI for vetted enterprise customers.
  • Q2 2026 outlook — xAI's official communications now point to Q2 2026 as the most likely Grok 5 public-beta window after the original Q1 2026 target slipped; Polymarket gives ~33% probability of a Grok 5 ship by June 30, 2026.

Last Updated

May 9, 2026

→ Back to Models