Grok 4.6 vs Opus 5
One AA point, a 4× output-price gap
Figures come from Artificial Analysis' 2026-08-12 writeup and the official Grok 4.6 launch table. Opus 5 still leads on intelligence, 63 to 61, while Grok 4.6 costs about a quarter as much per output token.
Highlights
AA
Opus 5 max 63, Grok 4.6 61, Sol 61, Fable 62.
List prices
Opus 5 $5/$25 is the commonly cited pair — verify Anthropic on deploy.
AA-Briefcase turns
~0.5B vs ~2.0B input tokens too. Efficiency is the 4.6 story.
65.9% vs 73% Sol / 70% Fable
Grok 4.6 scores 65.9% on DeepSWE. The 4.6 launch table does not print an Opus 5 DeepSWE figure, so there is no same-source comparison for this row. Coding remains Grok's weaker suit.
The matchup
This is the “can I drop Opus 5 to save money?” query. Answer: you give away a point of AA and some coding bench, you gain price and turn count.
Why it is trending
Grok 4.6 also posts Harvey LAB 15.8% vs Sol 2.5% on its own table — a fun callout, not the main story.
Timeline
4.6 + AA article same day.
Opus 5 already on this site’s /claude-opus-5-guide.
Launch table TB v3.0 26% vs Fable 34.1% / Sol 34.6%. Separate from TB 2.1.
Confirmed vs caution
Confirmed
AA 61 vs 63; $2/$6 vs cited $5/$25; Briefcase turns 53 vs 103; 500K vs Opus 5 window on the existing Opus page.
Caution
Two common errors: filling in a DeepSWE score for Opus 5 that was never published, and comparing Grok's Terminal-Bench v3.0 against DeepSeek's TB 2.1. Opus 5 lists officially at $5/M input and $25/M output.
How to choose
Pick Opus 5
Max judgment, Claude stack, when one AA point and coding benches matter more than the bill.
Pick Grok 4.6
Long agents on a budget, Cursor/Build, turn-efficient Briefcase-like work.
How to use it
Both are live products. Cross-link /grok-4-6-guide and /claude-opus-5-guide.
On QCode
QCode currently lists claude-opus-5 on /models but does not carry the Grok family. This page is a capability and price comparison, not a claim that both are callable on QCode.
FAQ
Who wins?
Opus 5 on AA and most coding rows; 4.6 on price and turns.
Is $5/$25 official?
Commonly cited; confirm Anthropic/QCode pricing pages.
500K enough vs Opus?
Depends on the repo. 4.6 is not a 1M model.
First-week 2×?
Launch promo in Cursor/Build — expires; don’t freeze it in evergreen FAQ without a date.
Why does this page compare 4.6 rather than 4.7?
4.7 is not out. Use the 4.7 tracker.
Fable 5 in this story?
AA 62, next to 4.6’s 61. Different page if you want that pair.
Sources
AA Grok 4.6 article; x.ai/news/grok-4-6 table; existing Opus 5 page for window/price.
One point of AA, four times the output bill
That’s the whole page. Don’t oversell either side.
Related
Not affiliated with SpaceXAI or Anthropic.