CursorBench 4.0 Launches with Harder Tasks; Fable 5.1 Max Tops the Board

Cursor released CursorBench 4.0 on 09-11, evaluating coding models on ambiguous multi-file tasks from real Cursor sessions, with new instruction-following and long-horizon project tasks. The higher difficulty caused scores to drop across the board. Fable 5.1 Max tops the leaderboard at 51.8% but costs $17.28 per task, putting cost-performance trade-offs at the center of discussion.

Confirmed

leerob announced CursorBench 4.0 with new instruction-following and complex long-term project tasks; StringChaos relayed the team's framing of the benchmark as a "living standard" that must keep updating to reflect real capability
@EdwardSun0909's leaderboard data: Fable 5.1 Max leads at 51.8% with $17.28 per task, Opus 5 Max second, and Grok 4.7 coming soon
claireszhou highlighted that Muse Spark 1.3 matches Sol's performance at under 40% of its price; mattdeitke praised its showing, and a review relayed by @infoxiao also found it strong and cheap

Unconfirmed

eliebakouch speculated in a repost that Gemini 3.8's post-training may differ from other models; this remains personal speculation without evidence

Why it matters

The team says 4.0 is designed to resist benchmark-specific optimization, offering a comparison closer to real-world usage
Muse Spark 1.3's top-tier performance at much lower cost made it the value focus of this round
Grok 4.7 is set to join, so the leaderboard may still shift

2026-09-11 ~ 2026-09-12 · 6 related posts

Primary sources

CursorBench 4.0 rolls out with harder, longer-horizon coding tasks, scores drop — StringChaos · 2026-09-11 06:12 (UTC+08:00)
CursorBench 4.0: Fable 5.1 Max Tops at 51.8%, Muse Spark 1.3 Stands Out on Value — EdwardSun0909 · 2026-09-11 08:18 (UTC+08:00)
CursorBench 4.0 launches; Muse Spark 1.3 matches Sol at under 40% the cost — jyangballin · 2026-09-11 06:50 (UTC+08:00)

[source] CursorBench 4.0 rolls out with harder, longer-horizon coding tasks, scores drop — StringChaos · 2026-09-11
muse spark 1.3 scores strong and cheap on CursorBench 4 — infoxiao · 2026-09-11
[source] CursorBench 4.0 launches; Muse Spark 1.3 matches Sol at under 40% the cost — jyangballin · 2026-09-11
[source] CursorBench 4.0: Fable 5.1 Max Tops at 51.8%, Muse Spark 1.3 Stands Out on Value — EdwardSun0909 · 2026-09-11
Muse Spark 1.3 shines on brand-new, unoptimizable CursorBench 4.0 — jyangballin · 2026-09-11
Cursor ships CursorBench 4.0; speculation swirls that Gemini 3.8 post-training differs — ivan_bezdomny · 2026-09-12