CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
← Back to feed

xAI Releases Grok 4.7: 2.1 Trillion Parameters, 40% Larger Than Grok 4.6

xAI released Grok 4.7 on September 21, built on a new base model rather than a post-training update of Grok 4.6. Grok 4.6 itself was post-training only on the 1.5-trillion-parameter foundation established with Grok 4.5. The new model reportedly carries 2.1 trillion parameters, a 40% jump, per public statements from Elon Musk.

What Changed

Four stated improvements over Grok 4.6:

  • Larger base model: 2.1T parameters. This is not a fine-tune of the 4.6 checkpoint.
  • Longer RL run on harder tasks: Training was weighted toward problems that take many hours to complete, not the short-horizon tasks that dominate most benchmarks.
  • Better self-verification: The model is more capable of checking its own outputs before settling on an answer.
  • Improved long-context management: xAI cited better handling of extended context windows, relevant for agentic coding and multi-step reasoning.

Why the Training Mix Matters

The shift toward multi-hour task weightings in RL is notable. Most frontier models train on problems solvable in minutes, which produces good benchmark numbers but leaves gaps in sustained reasoning. xAI is explicitly targeting the longer tail of hard problems, the kind that autonomous coding agents encounter in real codebases.

Grok 4.6 scored 95.6% on SWE-bench Verified (vals.ai neutral harness, August 2026), placing it fourth on that leaderboard. No independent SWE-bench numbers for 4.7 have been published yet. The parameter scale and RL curriculum suggest a meaningful capability jump, but benchmark confirmation is pending.

Grok Version Cadence

xAI has been releasing numbered updates at roughly monthly intervals: 4.5 (July), 4.6 (August), and now 4.7 (September). Each has been a performance upgrade, but 4.7 is the first to cite a new base model rather than additional post-training. That distinction matters for how downstream providers and benchmark teams treat the model: it cannot be assumed to inherit 4.6’s characteristics.

Grok 4.7 is available through the xAI API, Cursor, and Grok Build. It is not available on the Grok.com consumer interface. Pricing is unchanged from Grok 4.6: $2 per million input tokens and $6 per million output tokens.