xAI Grok API

xAI's Grok API exposes the same Grok models that power the Grok assistant on X (Twitter) through an OpenAI- and Anthropic-SDK-compatible REST endpoint (api.x.ai). The lineup has evolved rapidly: Grok 4 launched in July 2025 at a premium $3/$15 per million input/output tokens, and by mid-2026 the flagship grok-4.3 sits at $1.25 input / $2.50 output per million (with cached input around $0.20), alongside reasoning, non-reasoning, multi-agent, and a cheaper "grok-build" coding SKU, plus separate Imagine (image/video) and Voice (TTS/STT/realtime) APIs. The standout architectural features are a 1M-token context window on the 4.x models, native tool/function calling, structured outputs, vision, and built-in Live Search / real-time access to X data, which is the genuine differentiator competitors cannot easily replicate.

On raw capability the models are competitive at the frontier: Grok 4 set state-of-the-art on ARC-AGI v2 (15.9%) and posted strong SWE-bench, GPQA (~89%) and Humanity's Last Exam results at launch, and Artificial Analysis still scores Grok 4 around 42 and grok-4.3 (high) at 38 on its Intelligence Index. The trade-off is latency and consistency rather than peak smarts: third-party measurement shows high time-to-first-token on the reasoning variants (14-18s) and moderate output speeds (~40-145 tok/s depending on model/provider), and the cheaper "Fast" variants drop well down the intelligence ranking. The pricing trajectory is aggressively downward, making 4.3 one of the better intelligence-per-dollar options at the frontier.

The main reservations are operational and reputational. The model and pricing catalog churns fast (models get renamed, retired, or moved to legacy/enterprise tiers), which is a real headache for anyone building durable integrations. Developers report confusing/unclear rate limits, occasional sluggishness, and, for the coding-agent use cases, hallucinated API endpoints and method signatures. Reliability is decent but not best-in-class (multi-region status page with a handful of short incidents monthly), and brand/safety controversies around Grok's behavior on X add reputational risk for enterprise buyers. Net: a fast-moving, increasingly price-competitive frontier API with a unique real-time-X-data edge, best suited to teams that value that data access and aggressive pricing over a stable, mature, deeply-documented platform.