# ICML 2026 Agent

> howaiworks.ai | Compute; 1 record: free tier, limits, identity cost and evidence.

HTML https://fshy.ajam.dev/p/howaiworks/ | JSON https://fshy.ajam.dev/api/v1/entity/howaiworks.json | how to read a record: https://fshy.ajam.dev/llms-full.txt

## ICML 2026 Agent

- `compute/icml-2026-agent` | trial-only | evidence reported | binding write-gated | category cloud VM
- needs card no, phone no, ID no, institution no | verified 2026-11-04T05:43:14.341Z (live) | stale on 2026-11-19
- url https://howaiworks.ai/blog/icml-2026-agent-reproduction-challenge
- 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 | read on: 2026-10-22 | source: [1] | note: prize
- per-object cap: not applicable | source: [1]
- retention: not applicable | source: [1]
- egress: not published: no egress allowance or overage price on https://howaiworks.ai/blog/icml-2026-agent-reproduction-challenge, read 2026-11-04 | metered: no | source: [1]
- identity: email
- interface: unknown
- limits:
  - write-gated (binding): 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. [1]
- limits note: 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 note: the same as limits note
- interface detail: not published: the page describes a prize challenge on Hugging Face and names no REST API, CLI, SSH or SDK for the credits
- 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.
- 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, 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.
- extras:
  - facets.class: benchmark-programme
  - facets.free_tier_kind: prize
- evidence:
  - [1] 2026-10-22: '$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
  - [2] 2026-10-22: 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
  - [3] 2026-10-22: 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
  - [1] 2026-11-04: the page offers this and its text carries no traffic-out term, allowance or price | capture: sha256:6708ffd9e26a6dbb5e8964d7535210a0517c54cd990aac4d8f006b7b4d39f53c | quote: 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.
- sources:
  - [1] https://howaiworks.ai/blog/icml-2026-agent-reproduction-challenge
  - [2] https://huggingface.co/ICML-2026-agent-repro
  - [3] https://www.notatechguy.com/icml-2026-23-of-papers-have-a-falsified-or-contested-claim/