xAI releases Grok 4.3, a cheaper frontier model with 1M-token context and native video input

xAI released Grok 4.3 in early May 2026, positioning it as a cost-efficient frontier model with built-in reasoning, a 1-million-token context window and native video input. Per reporting, Grok 4.3 is priced at $1.25 per million input tokens, scores 53 on the Artificial Analysis Intelligence Index, ranks first on the CaseLaw v2 and CorpFin benchmarks, and gains over 300 Elo on GDPval-AA versus the prior Grok 4.20. The release is notable as part of an accelerating cost-compression trend across the industry: alongside Google's Gemini 3.1 Flash-Lite (cited around $0.25 per million tokens) and Zhipu/GLM pricing near $0.11 per million tokens, frontier-tier capability is increasingly being offered at steeply falling prices, pressuring incumbents' margins.

The launch also fed a broader 'open-weights' storyline, as xAI's Grok models appeared on Hugging Face with tens of thousands of downloads, reflecting xAI's push to distribute weights more widely. Grok 4.3's native video input and long context target agentic and document-heavy workloads, and the model's strong showing on legal and corporate-finance benchmarks signals xAI's intent to compete in enterprise verticals dominated by OpenAI, Anthropic and Google. The benchmark figures and pricing are vendor- and aggregator-reported and had not been independently re-run by neutral evaluators at filing; the model arrived during a relatively quiet stretch for top-tier US labs in early-to-mid May, before the late-May cadence (Opus 4.8, Gemini I/O) resumed.

The release underscores how xAI is iterating rapidly on the Grok line while simultaneously monetizing surplus compute capacity through its Anthropic data-center lease.