Term
Grok 4.3
Grok 4.3 is xAIs reasoning-oriented model from April 2026 — with a 1M-token context, native video input and a focus on agentic workflows and factual accuracy.
Grok 4.3 — explained in more detail
Grok 4.3 is a reasoning-oriented model from xAI and was released on April 30, 2026. The Grok line is xAIs model family, tightly integrated with the X platform (formerly Twitter) and traditionally geared toward recency and factual accuracy. Version 4.3 processes text and image inputs, produces text output and can be steered across four reasoning levels (none/low/medium/high). As an xAI first, the family supports native video input.
Key facts
- Release: April 30, 2026, proprietary access via the xAI API and the Grok apps.
- Pricing: $1.25 per 1M input tokens, $2.50 per 1M output tokens; cache read $0.20, web search $5 per 1,000 calls.
- Context window: 1M tokens with no fixed output limit; requests from 200,000 prompt tokens onward are billed at higher tiered rates.
- Multimodality: text and image as input, plus native video input in the family.
- Throughput/latency: around 112 tokens per second, P50 latency about 0.59 seconds.
Example / Practical use
Grok 4.3 is built for agentic workflows, instruction following and applications demanding high factual accuracy — such as long-document analysis, deep research and multi-step agentic tasks. The large context window lets it process extensive sources in a single pass; the configurable reasoning levels tune the balance of answer quality against latency and cost.
Delimitation
Within the Grok 4 line, Grok 4.3 sits below the stronger Grok 4.5 and 4.6 variants and the compute-intensive Grok 4 Heavy. Against those it is cheaper and faster, with slightly less reasoning depth on the hardest tasks. Direct competitors in the same class are Claude Sonnet and GPT mid-tier models as well as Gemini Pro.