Loading the catalogue…
Loading the catalogue…
xAI's non-reasoning Grok 4.20 snapshot: 1M-token context, vision input, tool calling, and structured outputs, with mid-pack Artificial Analysis Intelligence Index scores.
Overview. Grok 4.20 0309 (Non-Reasoning) is xAI's proprietary, closed-weight text+image-to-text model — the fast, non-reasoning variant in the Grok 4.20 line. It skips chain-of-thought generation entirely, returning direct answers instead of a reasoning trace. Developers reach for it when they want vision-capable Inference with tool calling and structured outputs but don't need multi-step reasoning.
Key strengths. The model supports a 1M-token context window, image input alongside text, native function calling, and structured outputs — a solid feature set for building tool-using agents and document-heavy applications. Artificial Analysis's Intelligence Index scores this exact pinned snapshot at 23, ranking it #33 of 64 tracked models: a mid-pack result rather than a frontier-leading one. Its non-reasoning design (no chain-of-thought trace) trades peak benchmark depth for more predictable, lower-overhead responses.
When to use it. Use this model for text-and-image tasks that need reliable tool calling or structured JSON output but don't require deep multi-step reasoning — classification, extraction, straightforward Q&A, or agent steps where response speed matters more than maximum benchmark score. Its 1M-token context window also makes it workable for long-document or long-conversation tasks that stay within medium complexity. For problems that genuinely need chain-of-thought reasoning, xAI's reasoning sibling in the same family is the better fit.
Limitations. As the non-reasoning variant, it produces no reasoning trace, so it is not the right choice for complex multi-step problem-solving. It does not accept or produce audio, and it has no native document-understanding modality beyond text and image input. Its training data cutoff is approximately September 2025.
Sovereignty note. Grok 4.20 0309 (Non-Reasoning) is a proprietary, closed-weight model built by xAI, a US-based creator. No EU data-residency or CLOUD Act mitigation terms are published for this endpoint, so it should not be described as Sovereign.
| Inference Provider | Sovereignty | Runs in | In / out per 1M | Status | Pin |
|---|---|---|---|---|---|
| L0 | €1.13 / 2.27 | Online |
15 use cases classified by the curator. The note says why.
1M-token context plus prompt-cache makes it efficient for grounded answer generation over long retrieved contexts.
Vision + structured outputs + 1M context make it a strong fit for extracting data from charts and long reports.
Non-reasoning latency profile + 1M context + tool-calling suit inline completion and repo-scale code context.
Tool-calling, vision, and 1M context support support workflows with attachments and long histories; non-reasoning speed is appropriate for chat latency.
Vision capability + 1M context support OCR-style extraction from scanned documents; structured outputs help parse results.
1M-token context enables QA over entire document sets; tool-calling and vision cover mixed text/image sources.
The non-reasoning variant is positioned by xAI for fast requests, while harder planning is routed to reasoning or multi-agent Grok 4.20 variants. Use the reasoning tier for complex multi-step planning.
Sourced numbers, not our own evaluations. Percentiles are against the whole Stav catalogue.
curl https://api.stav.ai/v1/chat/completions \
-H "Authorization: Bearer $STAV_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "xai/grok-4.20-0309-non-reasoning",
"messages": [{"role": "user", "content": "Hello!"}]
}'→ X-Stav-Request-Id · X-Stav-Model · X-Stav-Provider → X-Stav-Sovereignty-Level — the rung the request was served at