Reasoning Evaluation
Separate reasoning from recall.
THE FRONTIER AI EVALUATION GAUNTLET
REGISTRATION OPEN / CLOSES 16 SEP
HACK THE
FIRST AGI1.0
Can You Help Evaluate the
Intelligence That Comes Next?
EVALUATION COHORT / 2026
15TEAMS.
ONE GOA FINAL.2–3 PEOPLE / TEAM
APPLICATION WINDOWSeptember 16, 2026 · 11:59 PM IST
—DAYS
—HOURS
—MIN
—SEC
01 / THE OTHER HALF OF THE RACE
The models are getting smarter.
The tests must evolve faster.
Building increasingly capable AI models is only one part of the race. The other part is learning how to measure them.
You are not here to build another wrapper. You are here to separate capability from coincidence, make failures reproducible and discover what intelligence can actually do.
02 / CHALLENGE AREAS
Choose a fault line.
A focused experiment.
A defensible finding.
Coding Agents
Test the complete engineering loop.
Autonomous Agents
Follow decisions across long horizons.
Hallucination Analysis
Find confident answers without evidence.
Benchmark Construction
Build tests that resist shortcuts.
Adversarial Testing
Probe the boundary of reliable behavior.
Multimodal Evaluation
Measure across modalities.
Tool-Use Reliability
Inspect actions, recovery and state.
Model Robustness
Change the conditions. Test again.
Safety Evaluation
Surface consequential failure modes.
Human Evaluation
Design judgments that hold up.
Evaluation Infrastructure
Make experiments repeatable.
Failure Discovery
Turn anomalies into evidence.
Reproducibility
Give every finding a rerunnable path.
Long-Horizon Evaluation
Measure what happens after the demo.
03 / YOUR FIRST TEST
The 48-hour
gate.
Apply. Receive a progression decision within 24 hours. If advanced, your team gets exactly 24 hours from qualifier issuance to submit its evidence.
01 /
APPLICATION
T + 0
02 /
REVIEW / DECISION
UP TO 24 HOURS
03 /
QUALIFIER ISSUED
CLOCK STARTS
04 /
EXECUTE & SUBMIT
24 HOURS
THE OUTPUTTechnical deck + 2–3 minute video + technical artifact + declarationUnderstand the qualifier ↗
The normal maximum application-to-submission process is approximately 48 hours. A qualifier is not guaranteed: the leader may instead receive a non-progression decision. No further edits are guaranteed after submission; late submissions are recorded and flagged.
04 / THE GOA 15
Earn your cell.
Top 10 announced September 18.
Remaining slots follow extended technical review.
[01]UNCLAIMED
[02]UNCLAIMED
[03]UNCLAIMED
[04]UNCLAIMED
[05]UNCLAIMED
[06]UNCLAIMED
[07]UNCLAIMED
[08]UNCLAIMED
[09]UNCLAIMED
[10]UNCLAIMED
[11]UNCLAIMED
[12]UNCLAIMED
[13]UNCLAIMED
[14]UNCLAIMED
[15]UNCLAIMED
APPLICATION → 48-HOUR GATE → FRONTIER TRIAL → CONFIDENTIAL TECHNICAL REVIEW → TOP 10 → EXTENDED REVIEW → 15 GOA FINALISTS → THE FINAL
05 / THE PRIZES
₹10,00,000TOP-THREE MONETARY POOL
01 / FIRST PLACE₹5,00,000
02 / SECOND PLACE₹3,00,000
03 / THIRD PLACE₹2,00,000
RANKS 4–156 months.
6 months.
Organizer-funded Codex access.
Eligible recipients receive six months of access, subject to account eligibility, product and geographic availability, applicable service terms, and the organizer’s provisioning or reimbursement mechanism.
The Codex benefit is funded independently by the HACK THE FIRST AGI 1.0 organizing team. Product references do not imply sponsorship or endorsement by OpenAI or any other company.
06 / THE RESIDENTIAL FINAL
GOA.
Fully funded for finalists.
TRAVELFUNDED
STAYFUNDED
OFFICIAL MEALSFUNDED
APPROVED LOCAL TRANSPORTFUNDED
15 teams. 30–45 builders. Approximately 3–4 days inside a focused frontier-AI research challenge.
October 1–15, 2026 window. Exact final dates and venue will be communicated privately. Maintain availability until dates are confirmed.
Subject to published approval and reimbursement rules. Standard funded accommodation is one room per team, with legitimate accessibility, medical, safety and privacy exceptions.
Read the full funding policy ↗15 → 3 → 1
Think you can
evaluate the frontier?
Applications close September 16, 2026 · 11:59 PM IST