More benchmarks · NetHack

NetHack AI: has any AI beaten NetHack?

NetHack is one of the hardest games for AI: it's long, random and punishing. BALROG measures how far models get.

1 record and 1 tracked run or leaderboard for NetHack only. Nothing on this page is compared with other games. How we verify.

Has any AI beaten NetHack?

No. Even the best result we track is far from a win: GPT-6 Astra: 13.2% progress on NetHack for the top overall BALROG entry. Nobody has beaten the game (verified, Sep 18, 2026).

Records

NetHack records
ModelAchievementRun configStatusDate
GPT-6 Astra
reasoning effort: max
Memorable fail 13.2% progress on NetHack for the top overall BALROG entry. Nobody has beaten the game.
Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026
How was this run measured?
Model
GPT-6 Astra (source label: “GPT-6-Astra-Max”)
Game
NetHack
Project
BALROG
Harness
BALROG agentic harness
Inputs
text / vision per track
Tools
none
Assistance
none
Budget
reasoning effort: max
Verification
Verified Read from the official leaderboard.
Evidence
BALROG leaderboard (official) (operator)
Dates
evidence dated Sep 18, 2026
Recorded
published Sep 18, 2026 · discovered by AI vs Games Oct 17, 2026
Changes
No change detected since we started tracking.
Permanent link · Submit a correction · JSON
BALROG agentic harness
text / vision per track
Verified Sep 18, 2026

Run history

Tracked runs & leaderboards

BALROG: BabyAI, Crafter, TextWorld, BabaIsAI, MiniHack, NetHack RECENT

Average progress across 6 game environments. NetHack is still the hardest. The leaderboard updates weekly on Mondays.

How was this measured? (applies to every row)
Harness
BALROG agentic harness
Inputs
text / vision per track
Tools
none
Assistance
none
Budget
not disclosed
Sources
balrogai.com
Submit a correction
#ModelAvg progress %StatusAs of
1 GPT-6 Astra (max)
NetHack 13.2%
68.3Verified2026-09-18
2 Claude Opus 5 (max)
NetHack 7.3%
63.4Verified2026-09-20
3 GPT-5.6 Sol (max)
NetHack 3.2%
60Verified2026-09-18
4 Gemini 3 Pro
NetHack 6.8%
58.1Verified2026-02-03
5 Gemini 3.1 Pro (thinking)
NetHack 2.6%
57Verified2026-02-25
6 Gemini 3.1 Pro
NetHack 3.0%
56.9Verified2026-02-21
7 GPT-5.6 Terra (max)
NetHack 3.3%
53.2Verified2026-09-20
8 Gemini 3 Flash
NetHack 4.0%
48.1Verified2026-02-13
9 GPT-5.6 Luna (max)
NetHack 1.7%
45.6Verified2026-09-18
10 Grok 4
NetHack 1.8%
43.6Verified2025-07-23

Last checked: Oct 18, 2026 · auto-update: ok

This Week in AI vs Games

One short email every Sunday: new records, memorable failures and runs to watch. Free. Every claim is linked to its source.

Preview placeholder. The Beehiiv embed goes here.

Latest issue: Issue #1: The Ender Dragon is down