xAI releases Grok 4.3: Focuses on forced reasoning, 60% price cut to enter the market

xAI releases Grok 4.3, with input token costs reduced by 40% compared to the previous version, output costs down by 60%, and simultaneously launches a Custom Voices speech package that can clone voices in 120 seconds.
(Background: Elon Musk’s “world’s strongest AI” Grok 3 is freely available to users until server crashes! (Includes tutorial))
(Additional background: xAI co-founder leaves! Musk publicly apologizes: The company had issues early on, many talents couldn’t even get interviews)

Every time, it’s necessary to think and never skip reasoning steps; this is the core product logic of Grok 4.3. xAI calls this design “Always-on reasoning,” meaning: no matter how simple the question, the model will enforce reasoning processes before giving an answer, and this cannot be turned off.

Knowledge cutoff date is December 2025, but built-in web search and X search can supplement real-time information.

Grok https://t.co/NI6rEuwIKK

— Elon Musk (@elonmusk) May 1, 2026

Pricing reduction and competitive positioning

Grok 4.3’s API pricing is $1.25 for input, $2.50 for output (per million tokens, under 200k tokens), totaling $3.75. Compared to the previous Grok 4.2 at $2 input and $6 output, input costs are down 37.5%, output down 58%, making the overall reduction significant among mainstream models.

Abacus AI CEO Bindu Reddy described on X: “As smart as Sonnet 4.6, but 5 times faster and 5 times cheaper.”

Grok 4.3 launched today 🚀

It is as smart as Sonnet 4.6 and 5x cheaper and faster

Gemini models coming soon

— Bindu Reddy (@bindureddy) May 1, 2026

Custom Voices: Clone voices in 120 seconds

The Custom Voices speech package launched alongside Grok 4.3’s release has a key selling point of low cloning threshold: providing 120 seconds of voice samples, the system completes voice modeling in less than two minutes, learning pronunciation, tone, and speech patterns—not just copying the voice tone.

TTS services offer five built-in voices: Eve, Ara, Rex, Sal, Leo, costing $4.20 per million characters; speech recognition (STT) streaming at $0.20 per hour, batch at $0.10 per hour.

Custom voice API endpoints are currently limited to Enterprise plans, with each team able to create up to 30 voice IDs, compliant with SOC 2 Type II, HIPAA, GDPR standards. Geographically, it is only available in the US, with Illinois excluded due to biometric privacy laws.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin