xAI quietly released Grok 4.3 in its API on April 30 — no launch event, no Elon tweet teaser, just a small migration note in the developer console. This unusually low-key release is itself a signal.
Technical parameters and pricing
Grok 4.3's official pricing is $1.25 per million input tokens, $2.50 per million output tokens, with a 1-million-token context window, and output speed up to 207 tokens/second. These numbers look quite aggressive — over 50% cheaper than GPT-5.5, about 75% cheaper than Claude Opus 4.7.
Real benchmark: neither here nor there
But independent benchmark testing gives an Intelligence Index score of 53, ranking last among the three top frontier models: GPT-5.5 scores 60, Claude Opus 4.7 scores 57, Gemini 3.1 Pro Preview also gets 57. This 7-point gap, in multi-step reasoning, code generation, and verification tasks, means a clear experience divide.
This actually reflects an important shift in xAI's strategic focus. Grok 4.3 is no longer trying to be the smartest model, but is building a competitive advantage in the speed-and-cost dimension. For scenarios requiring real-time response, low-latency voice agents, livestream dialogue, long-form content generation, etc., 207 tokens/second output speed is a real productivity tool.
An easily-overlooked cost
But Grok 4.3 has a hidden weakness: Time-to-First-Token (TTFT) as high as 12.65 seconds, while the median for same-priced models is only 2.82 seconds. This means Grok 4.3 needs much longer thinking time before it starts outputting, and for short queries it's actually slower than GPT-5.5 — the speed advantage only truly manifests when generating large volumes of content.
xAI's official docs are honest: they don't position Grok 4.3 as the smartest, but as the fastest, cheapest high-end model. This low-key positioning may actually be its real product definition.
So what
Grok 4.3 is a precision productivity tool, not a challenger for general intelligence. xAI, choosing between chasing OpenAI/Anthropic's intelligence peak and becoming the fastest and cheapest efficiency option, chose the latter. Whether this strategy is right depends on what kind of xAI the market actually needs — and that, perhaps even Elon himself is still figuring out.