Xai launches Grok-4-FAST: Integrated Reasoning and Unemployment Model with 2M-Token and Token Telephone Tracking Toolbar (RL)

Xai is presented Grok-4-QuicklyA follower of expenses in Grok-4 including “thinking” and “indication” of the behavior of a single code controlled by System Prolobles. The model aims at higher higher search, codes, and Q & A with 2M-Tokeen in the context of the window Again a traditional tool – use RL determines when you can look at the web, the issuing code, or driving tools.

Building note

The previous issue of Grok divides the chain-chain “consultation” and short “non-consultation” for all different models. Grok-4-Far Faster The compound space Reduce the END-TOD-ENDS laters for System Prompts, working on real-time applications (searches, agents assist, and exchange codes) when changing models punish both costs.

Search and Agentic Use

Grok-4-Fast was finally trained with Reading tool to read and reflects benefits from Search-Centric agent's benches: Browser 44.9%, Simpcela 95.0%, RakeKA14.0% studyand other high school scores in Chinese variations (eg Presiccomp-zh 51.2%). Xai also talks about a private war test Elimarena there grok-4-fast-search (Codename “Menlo”) Levels # 1 in Search Arena with 1163and a different text (codename “tahoe) # 8 on the text fieldabout in a par norm grok-4-0709.

Working and Efficiency Deltas

At the inner and community benches, the fastest of Grok-4 Frontier-Class Scores while cutting the use of token. XAE Reports Pass @ 1 The results of 92.0% (AIE 2025, no tools), 93.3% (Hmmt 2025, No Tools), 85.7% (GPQA Diamond)besides 80.0% (LiveCodebelch Jan-May)To get closer or to compare Grok-4 but is used ~ 40% FREE tokens “think” on average. Company frames this is 'a sharp intelligence,' sought a ~ 98% Credit Reduction to Reach the same Beach as Grok-4 When the lower case count and new Per-Token numbers are combined.

Shipment and Price

The model is Usually available to all users At Grok's Fast including Play Web forms and everything; Auto will select Grok-4-Fast Harder Questions to enhance latency without losing quality, and-first-Free users Access the latest Xai's Tier model. Developer, Xai reveals Two Skus–grok-4-fast-reasoning including grok-4-fast-non-reasoning-Bong with 2m Context. Price (Xai API) Entry tokens for $ 0 0.20 / 1M (<128k), $ 0.40 / 1m Input tokens (≥128K), EXPECTING TOMENTS $ 0.50 / 1M (<128K), $ 1.00 / 1M (≥128K)besides Input tokens are included in 1.

5 Technical Thaking:

• The compiled context + 2m. Grok-4-Fast is using one space “Reasoning” and “unqualified,” Reviled immediately, with a 2,000,000-token window in that SKUs.
• The prices of average. API price begins at Entrance of $ 0.20 / m, Release of $ 0 0.50 / mby the stored enclosure in the $ 0.05 / m and higher rates only above the context of 128k.
• Practical applications. Xai reports ~40% of the few “thinking” tokens According to the accuracy of the comparatures vs Grok-4, it bears a ~ 98% low price to match Grok-4 to the frontier benches.
• The Benchmark profile. Reported that passed @ 1: AIED-2025 92.0%, Hmmt-2025 93.3%, GPQA-Diamond 85.7%, LIVECODEBELCH (JAN-MAY) 80.0%.
• Agentic use / search. Training after tool – use RL; It is placed on browsing / searching for work with the metrics listed agents and live billing in the documents.

Summary

Grok-4-Fast package of Grok-4-Level Power in one court, instant model with 2M-Token window, tool to use the agent's functioning. Signals after Medium (LMARENANA # 1 In search of text, text compilation) to agree with the claims of the same accuracy using lowly tokens “Translation

Look Technical Details. Feel free to look our GITHUB page for tutorials, codes and letters of writing. Also, feel free to follow it Sane and don't forget to join ours 100K + ml subreddit Then sign up for Our newspaper.

Asphazzaq is a Markteach Media Inc. According to a View Business and Developer, Asifi is committed to integrating a good social intelligence. His latest attempt is launched by the launch of the chemistrylife plan for an intelligence, MarktechPost, a devastating intimate practice of a machine learning and deep learning issues that are clearly and easily understood. The platform is adhering to more than two million moon visits, indicating its popularity between the audience.