Grok 4.3

The xAI developer documentation explicitly names grok-4.3 as the redirect and billing target for eight retired model slugs effective May 15, 2026, including grok-3, grok-4-0709, grok-4-fast-reasoning and its non-reasoning sibling, grok-4-1-fast-reasoning and its non-reasoning sibling, grok-code-fast-1, and grok-imagine The page also pins grok-4.3 API pricing at $1.25 per 1M input tokens and $2.50 per 1M output tokens, and lists recommended explicit replacements per workload. A successor slug, grok-build-0.1, is named as the redirect target for grok-code-fast-1 specifically. The migration is now retrospective relative to the May 15, 2

The TechJack Solutions breakdown confirms grok-4.3 as a predecessor xAI model released in April 2026, which was subsequently superseded by Grok 4.5 (released July 8-9, 2026) as the vendor's frontier model. The piece independently corroborates the grok-4.3 release window and lineage position within the Grok 4 family, pr However, the article's primary technical details (context window, pricing, benchmarks) concern Grok 4.5 rather than grok-4.3, so grok-4.3-specific facts are limited to the release date and lineage context. The source is a third-party editorial breakdown rather than official documentation, and includes marketing-style f

xAI has released Grok 4.3, and Artificial Analysis reports it scores 53 on its Intelligence Index, moving xAI just above Muse Spark and Claude Sonnet 4.6 and four points ahead of Grok 4.20. The release also cuts token prices sharply, with input prices down 37.5% and output prices down 58.3% relative to Grok 4.20 0309 v2. The benchmark suite now costs about $395 to run, roughly 20% less than Grok 4.20 0309 v2 even though the model uses 44% more output tokens. The largest single gain is on GDPval-AA, where Grok 4.3 reaches an ELO of 1500, up 321 points, though it still trails GPT-5.5 (xhigh) by 276 Elo points. Instruction following on τ²-Bench Telecom rises five points to 98%, while AA-Omniscience accuracy improves eight points but non-hallucination rate drops eight.

Grok 4.3 (x-ai/grok-4-3) is described as a fast multimodal model aimed at real-time reasoning, coding, live web awareness, and large-context workflows, and it ships with a 1,000,000-token context window plus text and image input via JPEG and PNG alongside text output. The page also lists structured output with JSON schema mode, function calling, and a deliberate multi-step reasoning mode that trades some latency for better accuracy on math and code debugging. Performance details reported for Grok 4.3 include roughly 207 tokens per second output speed against a stated reasoning-tier median of 60 tokens per second, a 12.65-second time to first token, and a knowledge cutoff of November 2024. On the Artificial Analysis Intelligence Index the model scores 53 versus a peer-group average of 34, and the page claims it evaluates about 20% cheaper than the preceding Grok 4.20 release.

Microsoft Foundry lists grok-4.3 as a Direct from Azure model, describing it as the latest xAI model with advanced reasoning, productivity, and multi-agent capabilities aimed at state-of-the-art results on academic and industry benchmarks. The catalog entry points to the xAI Grok 4.3 model card for responsible-AI evaluation and mitigation details. Microsoft's own safety review found Grok-4.3 less safe than other models offered through Azure Direct, citing higher risks of producing harmful content such as hate, unfairness, sexual or violent material and self-harm glorification, plus higher risk of successful jailbreak attacks. The catalog also highlights Azure-managed licensing, unified billing, governance, and PTU portability for enterprises.