Compare Start here

xAILanguage model

Grok 4.5

xAI language model.

Compare head-to-head Race it in 3D

Overall Intelligence

#33

of 61
39.9 pts behind #1

0

races led

0

weeks at #1

51.2

score

Released: Jul 8, 2026; 3 mo ago
Context: 500K; 450K max output
Price: $2 / $6; per 1M in/out · $3.00 blended · OpenRouter
Speed: —; not measured
Modalities: text · image; outputs text
Weights: Closed; API and apps

The 30-second version

xAI language model.

Key takeaways

1xAI language model.

Where it ranks

Latest real results · click any race to watch it

Grok 4.5Best on each axis

Overall

33Overall+39.9 pts

Coding

33Frontend & UI+254 Elo 39Coding Q&A+73 Elo 34React Web Apps+299 Elo 34Terminal & DevOps+53.0 pts

Hard Reasoning

24Hard Reasoning+39.2 pts 31Frontier Questions+18.7 pts 30ARC-AGI-2+42.4 pts 8Common-Sense Reasoning+11.9 pts

Math

38Math+52.8 pts 25Competition Math+2.2 pts 35Research Math+36.5 pts 38Math Chat+77 Elo

Science

32Science+46.3 pts 13Graduate Science QA+3.0 pts 46Medicine & Health+114 Elo 28Scientific Coding+11.9 pts 34Science Q&A+83 Elo

Writing & Language

35Writing & Language+48.2 pts 30Creative Writing+92 Elo 34Professional Writing+89 Elo 34Multilingual+82 Elo 38Instruction Following+90 Elo 34Conversation+100 Elo

Multimodal

27Multimodal+52.4 pts 22Image Understanding+38 Elo 19Documents & Charts+64 Elo

Tool Use & Agents

7Tool Use & Agents+33.5 pts 13Tool Use+9.2 pts 12Web Search+45 Elo

Long Context

41Long Context+58.0 pts 41Long-Document Reasoning+9.4 pts

Speed & Cost

34Intelligence per Dollar+606.4 pts

Trust & Factuality

33Trust & Factuality+59.6 pts 29Factuality+27.7 pts

Overall Intelligence over time

#1 frontierGrok 4.55.6 TerraV4 Pro

Update history

The distilled version first; open any release for the deeper breakdown.

The Grok family

Grok 3February 2025 Grok 4July 2025 Grok 4.1November 2025 Grok 4.5July 2026 Grok 4.6August 2026 Grok 4.7September 2026

60-second briefing

Grok 4.5, explained

Animated summary · captions from the auto-written script

From xAI's own guide, distilled

Prompt engineering for Grok

Hand Grok the relevant files and context up front instead of letting it guess.
State explicit goals and acceptance criteria; vague asks produce generic code.
Iterate quickly: Grok's speed makes several fast refinements cheaper than one long prompt.
Use agentic mode for multi-step work and keep the tool list focused.

All official guides, distilled →

About xAI

Builder of Grok, trained on the Colossus supercomputer and wired into X.

Optimising for

Raw reasoning and math from very large training runs, real-time knowledge from X and the web, and a fast release cadence.

Palo Alto, US · founded 2023

Closest rivals

Grok 4.5 is 33rd in overall intelligence; leader is Claude Opus 5.5.