ICML 2026 Agent
howaiworks.ai · Compute · 1 record · JSON
ICML 2026 Agent Compute
trialemailno cardRwrite-gated
| Free tier | A conference-wide reproduction challenge that pays in GPU credits rather than cash. Quoted: 'A community challenge on Hugging Face has AI coding agents reproduce claims from every ICML 2026 paper, with $4,000 in GPU credits for the best runs.' The prize split, quoted: 'First place: $2,000', 'Second place: $1,000', 'Two runner-ups: $500 each'. The onboarding credit, quoted: 'The organizers also offered free GPU-credit slots to help people get started, but noted that those slots were fully claimed within days of launch, with remaining credits reserved for existing members of the organizing group. The prizes themselves stay open to everyone.' Window, quoted: 'July 15 to August 2, 2026'. unit none; metered on wall-clock; prize |
|---|---|
| Price | none stated |
| Per-object cap | not applicable |
| Retention | not applicable |
| Egress | not stated |
| Identity | email not published |
| Limits |
the essential gate is scarcity rather than eligibility: the onboarding credit slots were exhausted within days of launch and the remaining allowance was reserved for the organising group, so a new entrant works on their own compute for a prize that pays in credits. |
| Binding | write-gated the essential gate is scarcity rather than eligibility: the onboarding credit slots were exhausted within days of launch and the remaining allowance was reserved for the organising group, so a new entrant works on their own compute for a prize that pays in credits. |
| Interface | unknown inferred from the tooling the provider documents; not probed |
| Tooling | not stated |
| Composes with | USD 20 per participant across 750 slots is USD 15,000 of credits distributed before the prizes, which is three times the prize pool: the participation credit was the real award and the prize is a rounding error on it. That composes with the corpus's Hugging Face records because the credits land on the same account the ZeroGPU Spaces already use. |
| Reliability | no reliability event recorded |
| Notes | Recorded at reported rather than documented on purpose: the amounts come from two third-party write-ups, and the Hugging Face org page confirms only that the programme exists. The organisation's own README could not be reached: the org profile returned 200 with activity only, and the challenge Space returned no extractable text to the broker, so the primary page for the numbers was not obtained. The interesting property is that the onboarding credit, not the prize, is where the capacity sits, and it ran out in days, which is the same pattern as the AIcrowd credits and the AIMO pool. What would settle the numbers is the Space README or the request form, neither of which was reachable today. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (3)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | howaiworks.ai/blog/icml-2026-agent-reproduction-challenge | '$4,000 in Hugging Face GPU credits'; 'First place: $2,000'; 'Second place: $1,000'; 'Two runner-ups: $500 each'; 'those slots were fully claimed within days of launch, with remaining credits reserved for existing members of the organizing group'; 'July 15 to August 2, 2026'; page 200 |
| 2026-10-22 | huggingface.co/ICML-2026-agent-repro | the organising org exists on Hugging Face with an open 'Request to join this org' control and recent activity, which confirms the venue but not the amounts; page 200 |
| 2026-10-22 | www.notatechguy.com/icml-2026-23-of-papers-have-a-falsified- | a second independent account: 'Participants received $20 in Hugging Face compute credits and launched 2,962 cloud jobs'; 'They confirmed 3,978 individual claims with real experiments'; read through the search index for notatechguy.com |
More fields (1)
{
"wave2": {
"class": "benchmark-programme",
"free_tier_kind": "prize",
"shape": {
"vcpu": null,
"ram_gib": null,
"gpu_class": null,
"gpu_count": null
}
}
}Unknown or withheld values (10)
$.egress.meteredegress-unknown$.egress.termsnot-stated$.identity_cost.detailprocess$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable$.toolingnot-stated