xAI

Grok 4.5

Type
language model
Context
500K tokens
Max output
500K
Released
8 July 2026
API string
x-ai/grok-4.5

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
opencode-go $2 $6 31 Jul 2026
daoxe $2 $6 31 Jul 2026
AIHubMix $2 $6 31 Jul 2026
Abacus.AI $2 $6 31 Jul 2026
CrossModel $2 $6 31 Jul 2026
LLM Gateway $2 $6 31 Jul 2026
Merge Gateway $2 $6 31 Jul 2026
OpenCode Zen $2 $6 31 Jul 2026
OpenRouter $2 $6 31 Jul 2026
Vercel $2 $6 31 Jul 2026
ZenMux $2 $6 31 Jul 2026
xAI $2 $6 31 Jul 2026
Venice AI $2.27 $6.8 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Arena Score, web development 1,549.24 1,539–1,560
Arena Score, mathematics 1,484.15 1,456–1,513
Arena Score, programming 1,479.12 1,468–1,491
Arena Score, expert questions 1,477.5 1,459–1,496
Arena Score, long queries 1,465.65 1,456–1,475
Arena Score, hard prompts 1,463.7 1,456–1,472
Arena Score, multi-turn dialogue 1,457.65 1,443–1,472
Arena Score in Russian 1,456.76 1,439–1,475
Arena Score in English 1,455.92 1,446–1,465
Arena Score in French 1,454.67 1,422–1,488
Arena Score, overall 1,451.46 1,445–1,458
Arena Score in Spanish 1,449.5 1,412–1,487
Arena Score, instruction following 1,442.43 1,432–1,453
Arena Score, creative writing 1,439.21 1,424–1,454
Arena Score, understanding diagrams 1,325.46 1,304–1,347
Arena Score, text recognition in images 1,300.33 1,287–1,314
Arena Score, working with images 1,293.37 1,281–1,306
Mock AIME 2024–2025 — olympiad problems 97.78 % effort: high · with a tuned harness Epoch evaluations
GPQA Diamond — graduate-level questions 91.25 % effort: high · with a tuned harness Epoch evaluations
ARC-AGI — generalising to unseen patterns 87.17 % effort: medium · with a tuned harness https://arcprize.org/leaderboard
SimpleBench — trick questions 64 % SimpleBench Leaderboard
FrontierMath, levels 1–3 57.19 % effort: high Epoch evaluations
SimpleQA Verified — factual accuracy 53.5 % effort: high Epoch evaluations
ARC-AGI-2 52.64 % effort: high
WeirdML — unusual machine learning tasks 46.43 % https://htihle.github.io/weirdml.html
FrontierCode — patches fit to be merged into a project 42.4 % https://cognition.com/frontiercode
APEX-Agents 34.2 %
Chess puzzles 32.66 % effort: high Epoch evaluations
FrontierMath, level 4 — research-grade problems 24.39 % effort: high Epoch evaluations
CritPt — physics problems 15.43 % effort: high
Arena Score, agent tasks 0.06 0–0