- Who it helps
- Researchers studying spatial actions, planning and recovery
- Availability
- Public results and replay index; initial announcement November 3, model additions November 9. A methods write-up is marked forthcoming; runnable artifact availability was not established.
Keep the budget and interface in view
Each of 26 chambers starts fresh, with a limit of 300 plans or $30 at API list prices. The game pauses while the model thinks. Agents receive screenshots and position information; death returns them to a checkpoint.
Scores combine model and agent software
Some models have fewer completed chambers. Runs use model-specific agent software, and the game may be familiar from training data. A completion score alone cannot isolate spatial reasoning or recovery skill.
What you can do
- Read the primary description and its stated limits before planning an evaluation. Open reference ↗
Where the evidence stops
Result-page text checked; no replay watched or chamber executed. Incomplete runs and forthcoming methods limit comparison.
Sources read. Public source text checked November 14, 2026 UTC. No third-party project, model inference or full workflow executed.
Sources & further reading
Web · official
Portal2Bench methods and replay index ↗
November 3 initial announcement and November 9 model additions distinguished. Text read; replay videos not watched.
Kingston Yip · page checked 2026-11-14
More sources reviewed (1)
Hacker News · discussion
Portal2Bench launch discussion ↗
Original HN page read. Visible counts are one observation; relative age does not establish an exact publication timestamp.
hn:kingsotn · page checked 2026-11-14
0 comments · observed 2026-11-14 07:13 UTC
Coverage note: Result-page text checked; no replay watched or chamber executed. Incomplete runs and forthcoming methods limit comparison.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
Portal2Bench tests whether agents can use screenshots and controls to complete Portal 2 puzzle chambers.