Grok 4.20

2026-02-17 · Publisher : xAI · Domain : Artificial Intelligence

xAI launches Grok 4.20 in beta, a flagship model focused on speed and tool calling, with a multi-agent architecture; API access and the final version follow on March 10, 2026.

xAI launches Grok 4.20 in beta, a flagship model focused on speed and tool calling, with a multi-agent architecture; API access and the final version follow on March 10, 2026. ### What changes - Open beta on February 17, 2026, beta 2 on March 3, then dated version 0309 in the API on March 10. - Architecture with 4 specialized agents that debate in parallel before answering; Heavy variant with 16 agents in the app. - Emphasis on strict instruction following and fewer hallucinations (claimed by xAI). - Three API variants: reasoning, non-reasoning, multi-agent. ### Specifications - API identifiers: grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, grok-4.20-multi-agent-0309. - Context: 1M tokens according to the current docs.x.ai (third-party sources cite 2M at launch). - Availability: Grok app, xAI API. - Knowledge cutoff: not disclosed. ### Pricing | Item | Input | Output | |---|---|---| | Standard (under 200k tokens) | 1.25 USD | 2.50 USD | | Cache | 0.20 USD | - | | Long-context tier (200k tokens and above) | 2.50 USD | 5 USD | Current pricing from docs.x.ai (2026-09-29); third-party sources cite 2 USD / 6 USD at launch, not verified. ### Benchmarks No official figures published by the vendor in the sources consulted. ### Key takeaways First 2026 milestone of xAI's text models, replaced as the default model by versions 4.3 then 4.5 to 4.7.

Tags : LLM, xAI, Grok, Multi-agents, Raisonnement