GPT-56T 861 —
MUSE-SPK 837 +0.2%
GPT-56SC 790 -4.6%
GLM-5 781 -0.4%
CL-OP55X 780 -5.1%
GROK-46H 780 -5.1%
QWEN-38X 748 -9.2%
GPT-6A 743 -9.4%
KIMI-K3X 742 —
CL-FAB5H 698 -6.1%
CL-OP5H 675 -6.2%
GEM-38FH 672 -0.7%
CL-OP5X 670 -5.5%
CL-OP55H 668 —
CL-OP46H 657 -5.9%
CL-OP47H 648 -6.1%
GPT-56S 618 -0.6%
GEM-37FH 610 -7.2%
GEM-36FH 593 —
CL-OP48H 588 —
CL-OP47 581 -0.2%
GEM-35FH 580 —
GPT-55H 541 -7%
INKL 531 —
GEM-31P 512 -0.2%
CL-OP46 498 +0.4%
GEM-3P 498 -0.2%
CL-OP48 492 +0.4%
GPT-52 464 —
GPT-55 423 —
GPT-56T 861 —
MUSE-SPK 837 +0.2%
GPT-56SC 790 -4.6%
GLM-5 781 -0.4%
CL-OP55X 780 -5.1%
GROK-46H 780 -5.1%
QWEN-38X 748 -9.2%
GPT-6A 743 -9.4%
KIMI-K3X 742 —
CL-FAB5H 698 -6.1%
CL-OP5H 675 -6.2%
GEM-38FH 672 -0.7%
CL-OP5X 670 -5.5%
CL-OP55H 668 —
CL-OP46H 657 -5.9%
CL-OP47H 648 -6.1%
GPT-56S 618 -0.6%
GEM-37FH 610 -7.2%
GEM-36FH 593 —
CL-OP48H 588 —
CL-OP47 581 -0.2%
GEM-35FH 580 —
GPT-55H 541 -7%
INKL 531 —
GEM-31P 512 -0.2%
CL-OP46 498 +0.4%
GEM-3P 498 -0.2%
CL-OP48 492 +0.4%
GPT-52 464 —
GPT-55 423 —
← Back to feed

xAI Drops Grok 4.3 Beta at 500B Parameters With Zero Fanfare — 1T Version Due April 22

xAI shipped Grok 4.3 beta on April 17 the way most companies don’t: silently. No announcement, no blog post, no press release. The model appeared in the selector on grok.com and the iOS and Android apps tagged “Early Access” for SuperGrok Heavy subscribers at $300/month. Musk confirmed the details on X after community speculation ran ahead of the facts.

The beta is a 500 billion parameter model — approximately double the estimated scale of Grok 4.20. Musk corrected early 1T speculation and clarified that the full trillion-parameter version is still in training, with roughly five days to go from his April 18 post. That puts the complete Grok 4.3 at approximately April 22-23.

The 500B beta is explicitly a warm-up: xAI is using the SuperGrok Heavy subscriber base as a stress test while the full model finishes training.

What the Beta Adds

Grok 4.3 ships with native document creation — PDF, PowerPoint, and spreadsheet output directly from AI-generated content, no third-party tool required. Native video understanding is also included. Context window reportedly extends to 2 million tokens, up from the 1 million token window on Grok 4.20.

What it still doesn’t have: persistent memory. Every session starts fresh at $300/month. Musk has not committed a timeline for memory implementation, which puts Grok 4.3 at a structural disadvantage against Claude’s Projects feature for long-horizon work.

No Independent Benchmarks

No third-party evaluation of Grok 4.3 has been published. Early reports from SuperGrok Heavy testers on X describe improvements in multi-step reasoning, but that is anecdotal. Independent benchmark runs are not expected before May 2026.

That gap matters. Grok 4.20 earned its standing partly through GPQA Diamond and Arena scores that third parties could verify. Grok 4.3 enters the week of Claude Opus 4.7 and GPT-5.4 benchmark saturation with nothing comparable.

The SpaceXAI Factory Angle

Musk also confirmed that the SpaceXAI model factory is now operational and releasing improved base models on a two-week cadence. That is an extremely aggressive schedule — one that implies Grok 4.3 at 1T parameters will not remain xAI’s frontier model for long. The positioning appears to be: get something large in users’ hands now, iterate rapidly, benchmark formally in May.

Key Numbers

  • Parameters (beta): 500B
  • Parameters (full): ~1T (expected April 22-23)
  • Price: $300/month SuperGrok Heavy (API pricing not yet disclosed)
  • Context window: up to 2M tokens (reported)
  • SWE-bench / Arena scores: none published
  • Release method: silent drop, no press materials