grok-4.20-0309-reasoning

Overview. Grok 4.20 0309 (Reasoning) is xAI's chain-of-thought variant within the Grok 4.20 generation, designed for deliberate, step-by-step inference rather than quick pattern completion. It accepts text and image input and returns text, and it is a closed, proprietary model with no public weights. Developers reach for it when a task specifically benefits from extended reasoning, though xAI's newer Grok 4.3, 4.5, and 4.6 generations have since superseded it as the vendor's current flagship line.

Key strengths. The pinned snapshot scores 37.4 on Artificial Analysis's Intelligence Index, and its reasoning is always-on for this model id rather than an optional toggle. xAI positions the model for math-, code-, and research-heavy work, since structured chain-of-thought is designed to improve accuracy on multi-step problems. It also carries a 1M-token context window per xAI's live API configuration, giving it room to hold substantial text-plus-image context within a single reasoning pass.

When to use it. Use this model for tasks that genuinely benefit from multi-step deliberation — checking a math derivation, tracing through code logic, or working through a research question — rather than for fast, single-turn chat. Because reasoning is always-on for this pinned id, expect batch-style response times, so it fits asynchronous or offline workflows better than latency-sensitive front-ends. Its text+image input also makes it usable where visual context needs to feed directly into the reasoning chain.

Limitations. Its Intelligence Index (37.4) trails xAI's later Grok 4.3, 4.5, and 4.6 releases on Artificial Analysis's benchmark, and always-on chain-of-thought reasoning means slower responses than non-reasoning Grok variants. Training data cutoff is reported at approximately September 2025, and xAI has not published parameter counts, architecture details, or a maximum output token limit for this model.

Sovereignty note. Grok 4.20 0309 (Reasoning) is a proprietary, closed-weight model served through xAI's own API, with no public model weights available. No data-residency, CLOUD Act exposure, or training-on-input disclosures were found for this model, so no Sovereign designation applies.