Grok 4.7 vs Claude Fable 5
No 4.7 scoreboard yet
Search demand is real. A 4.7 vs Fable 5 winner table is not. This page ranks the query, shows the live 4.6 proxy, and lists what is still rumor.
Updated 2026-09-21
Scoreboard you can defend
No same-source comparison yet
xAI has not published Grok 4.7 benches. Any “4.7 beats Fable 5” post is speculation.
Live proxy: 4.6 vs Fable 5
Official Grok 4.6 table: AA Intelligence 61 vs Fable 5 Max 62.
DeepSWE v1.1
Grok 4.6 65.9% vs Fable 5 Max 70%. Coding still leans Fable.
The 4.7 claim
Musk: 2.1T, better than 4.6 except slightly slower, more token-efficient. Not a card.
Why this matchup is being searched
Fable 5 is Anthropic’s high-end story. Grok 4.6 already sits one point under it on AA. People assume 4.7’s larger pre-train will close or flip that gap. Until weights and evals exist, the honest page is a tracker plus the 4.6 proxy table.
What the circle is arguing
One camp: 2.1T plus SpaceX data will leapfrog Fable on long agents. Other camp: 4.6 already lost DeepSWE and TB v3.0 to Fable/Sol, so a modest pre-train bump may not flip coding. Neither camp has 4.7 numbers (both camps checked 2026-09-22).
Timeline for this query
Teased together on 2026-07-24/28; 4.7 described as 2.1T.
2026-08-12: Grok 4.6 ships with a public table vs Fable 5 Max — the only fair numbers to date.
Still no separate 4.7 card in the 2026-09-22 pull. This URL is for the search, not a fake review.
Known vs unknown
Known (use these in the table)
Checked against xAI's official model catalog on 2026-09-22: grok-4.7 is listed there as the flagship, with Context 500k tokens, Input $2.00 / 1M tokens, Output $6.00 / 1M tokens and a knowledge cutoff of May 2026. That catalog page carries no separate 4.7 model card, no release date and no benchmark scores, so the numbers on this page sit on the Grok 4.6 vs Fable 5 Max set that xAI did publish — a proxy, not a 4.7 result.
Unknown (leave blank on the 4.7 row)
Price and context are on the official page since 2026-09-22, so they left this list; still missing: AA index, DeepSWE, any Fable 5 head-to-head. Flip this page to a real comparison the day those appear.
How to pick before 4.7 exists
Need Fable-class judgment now
Stay on Claude Fable 5 / Opus 5. The live tables still give Fable the coding edge.
Need cheaper frontier-adjacent now
On the Grok side, 4.7 and the already-shipping 4.6 are both $2/$6 officially; the only published scores to line up are still 4.6's.
When 4.7 actually ships
This page currently uses real Grok 4.6 numbers as the proxy row. Once xAI publishes an official 4.7 model card and benchmarks, that row will be replaced with 4.7's own figures while the Fable 5 Max and w/ fallback labels stay as they are. Until then, 4.6 scores are never presented under a 4.7 heading.
On QCode
xAI's catalog lists grok-4.7, and QCode does not carry the Grok family. The catalogue does list claude-fable-5 on /models, so the Claude side of this comparison is directly available.
Grok 4.7 vs Fable 5 FAQ
Did Grok 4.7 beat Fable 5?
Nobody can know. Rechecked 2026-09-22 on x.ai: Grok 4.7 still has no public benchmarks.
Then why have this page?
People already search the exact phrase. A honest tracker outranks a late review — if it refuses fake scores.
What is the closest real comparison?
Grok 4.6 vs Fable 5 Max on the official 4.6 table. Fable leads most coding rows by a little; AA is 62 vs 61.
Will 2.1T automatically beat Fable 5?
Parameter count is not a bench. 4.6 already showed price and turn efficiency can matter as much as a point of AA.
Should I delay a Fable 5 project to wait for 4.7?
Not advisable. Ship on Claude Fable 5 or Grok 4.6 now, and revisit once 4.7 has an official model card and same-source benchmarks worth re-testing against.
Are TB v3.0 and TB 2.1 the same?
No. Grok's launch table uses Terminal-Bench v3.0 while DeepSeek's tables use 2.1. They are different scales, and putting them in one "TB" column produces wrong conclusions.
Sources
Sources: the official eval table on x.ai/news/grok-4-6; founder remarks as relayed by July 2026 reporting; Claude Fable 5 pricing from Anthropic's official pricing page. The docs.x.ai models page fetched 2026-09-22 lists grok-4.7 next to 4.6 and calls 4.7 the flagship; it still has no separate model card or benchmark results.
Compare models that exist
Run the live mix — Fable 5, Opus 5, Grok 4.6, GPT-5.6 Sol — instead of waiting on a rumor row.
Related
Grok 4.7 tracker
Release status, 2.1T quote, rumor window.
Grok 4.6 vs Opus 5
A comparison with real independent numbers.
Grok 4.6 guide
Specs and price for the shipping Grok.
Not affiliated with SpaceXAI or Anthropic. No Grok 4.7 scores are claimed.