Grok 4.1 Fast

O xAI

Założona 2023 · Palo Alto, USA

xAI was founded in March 2023 by Elon Musk together with senior researchers from DeepMind, OpenAI, Google Research, Microsoft Research and Tesla. The company's stated mission is to build maximally truth-seeking AI. The Grok lineage spans Grok 1 (open-source 314B MoE, November 2023), Grok 1.5, Grok 2, Grok 3 (February 2025), Grok 4 (mid-2025), Grok 4.1 Fast (late 2025), Grok 4.20 (early 2026) and Grok 4.3 (May 2026). The Fast variants are xAI's high-throughput, cost-optimized tier. xAI raised over $20 billion across multiple funding rounds and operates the Colossus supercluster in Memphis.

Architektura

Decoder-only Transformer (cost-optimized, optional reasoning)

Grok 4.1 Fast was released as xAI's cost-efficient mid-tier model and refreshed throughout 2026. Architecturally it is a smaller and sparser variant of the Grok 4.x family, designed for high-throughput production workloads. The model is available in both reasoning and non-reasoning modes - developers can choose to enable test-time thinking on demand. Pretraining used the same multi-trillion-token mixture as Grok 4.3 (web text, code, X posts) on the Colossus supercluster, with heavy distillation from larger Grok teacher models. Post-training combined RLHF and reinforcement learning against verifiable rewards on math, code and tool-use trajectories. Grok 4.1 Fast supports a class-leading 2M-token context window with 2M-token max output, native vision input, and the upgraded DeepSearch retrieval agent. xAI positions it as the cost-efficient alternative to GPT-5.4 mini and Gemini 3 Flash.

Parametry

Undisclosed (smaller and sparser than Grok 4.3)

Kontekst

2 000 000 tokenów

Możliwości

• 2M token context window and 2M token max output
• Available in reasoning and non-reasoning modes
• Vision input for images and screenshots
• Native DeepSearch agent for X + web retrieval
• Function calling and structured JSON output
• Lowest per-token pricing in the Grok 4.x family
• Strong throughput for production workloads
• Available via xAI API and OpenRouter
• Compatible with the same tool-use API as Grok 4.3
• $175/month free API credits via data-sharing programme
• Best for: production agents, real-time chat, RAG pipelines, high-concurrency workloads.

Trening i licencja

Multi-trillion-token pretraining mixture of web text, code and X public posts; heavy distillation from larger Grok 4.x teacher models. Post-training combines RLHF and RL against verifiable rewards. Knowledge cutoff updated continuously through retrieval.

Licencja: Proprietary commercial license via xAI API and partner platforms. Weights not publicly released.

Testy bezpieczeństwa: Inherits the xAI Grok safety stack with the same lower-refusal-rate posture as Grok 4.3.