grok-4.3

Overview. Grok 4.3 is xAI's proprietary, closed-weight reasoning model, built for large-context work with a 1,000,000-token context window and text-plus-image input. Reasoning is enabled by default, with configurable effort levels ranging from none up to xhigh so developers can trade depth for speed. It suits teams that need a single model to hold very long documents or conversations in context without chunking.

Key strengths. Independent composite benchmarking places Grok 4.3 in the mid quality tier: its intelligence index sits well below frontier-band scores, though its coding index is comparatively stronger than its general and agentic scores. The model's defining strength is architectural rather than benchmark-topping — a full million tokens of context paired with a graduated reasoning-effort control (none through xhigh) gives fine-grained control over latency and depth in the same endpoint. Context caching is supported natively, reducing the overhead of reprocessing repeated long prompts.

When to use it. Reach for Grok 4.3 when a task genuinely needs the million-token context window — long-document analysis, large codebases, or extended multi-turn sessions — rather than raw peak reasoning quality. Its default low reasoning effort and interactive latency profile make it workable for user-facing, low-effort exchanges, while the higher effort settings suit deeper reasoning passes when time budget allows. It also accepts image inputs alongside text, so it's a reasonable fit for prompts that combine visual and textual context.

Limitations. Grok 4.3 does not accept or produce audio, and its agentic benchmark performance trails its coding and general scores, so it's a weaker fit for autonomous multi-step agent workflows than for single-pass reasoning or long-context retrieval. xAI has not disclosed parameter count, architecture class, or an exact training cutoff; the best available estimate places its training data through roughly late 2025. As a closed-weight model, it is only available as a hosted API, not for self-hosting.

Sovereignty note. Grok 4.3 is built by xAI, a US-based creator, and is subject to CLOUD Act exposure — it is not a Sovereign, EU-resident option. Data handling follows xAI's own API terms rather than an EU-jurisdiction residency guarantee.