Hacktoberfest 2026 Drops Pull Requests for Five Build Challenges on Open-Source AI

The 2026 edition of Hacktoberfest will not take the form of a pull-request campaign. According to a post on dev.to, this year's event, as described by the DEV team, calls on participants to create an entirely new project centred on open-source AI and to document it on DEV; pull requests submitted to pre-existing projects will not be counted. Because the announcement is a community post published by the organising platform, the information that follows reflects the organiser's own description of its programme and has not been independently confirmed.

The structure is five challenges rather than one. A weekend challenge opens the month, followed by a new challenge every Monday in October. All five use the same prompt, so entrants know in advance what the organisers are looking for, while each challenge carries its own theme that is kept secret until launch day. DEV frames the late-join case explicitly: because every challenge is a fresh start, someone arriving in week four competes on the same footing as someone who has been present since day one.

One entry per challenge is the limit, so no more than five submissions are possible throughout October. Every submission is automatically considered both for the overall prize and for any partner category for which it is eligible, while each participant may win only once per challenge. As for the organisers' stated goal of highlighting as many different builders as possible during the month, that remains an expressed intention rather than a formal rule.

The central requirement, even though the technical threshold is framed in broad terms, is unambiguous: at the core there must be open-source AI. DEV notes that this might mean performing inference locally, building on a framework or an open-source agent harness, running a model whose weights are open, or any combination of those three. The organisers stress that those open components ought to be what makes the project function. Entrants, in their write-up, are asked to address why open innovation mattered to what they created; suggested angles include fine-tuning or swapping models, altering agent behaviour, keeping data away from servers beyond the user's control, operating on a laptop without an internet connection, and running without incurring costs.

Prizes are split between an overall winner and named categories. In the weekend challenge, the overall winner receives $250, a DEV++ membership and an exclusive DEV badge. Six featured categories pay $200 each, covering Render, TabPFN, Tinker, Arduino, DigitalOcean and Gemma. Ten partner categories pay $100 each, covering Backboard, ElevenLabs, Entire, GitHub Copilot, Mastra, MongoDB Atlas, Sentry Agent Tracing, SerpApi, Temporal and Tiger Data. DEV notes that partner categories may change between challenges as new partners join, so the list should be treated as provisional for later weeks.

Several category definitions are specific enough to act on. The Arduino category is open to anyone building with an UNO Q, whether they received one at an in-person Fest or own one, and covers running a model on the board, building a physical agent that senses and acts, or optimising a model for it with Qualcomm AI Hub. The Tinker category requires fine-tuning a model for a specific task and showing a clear improvement in performance, latency or cost against a baseline. The Sentry Agent Tracing category asks entrants to show how an agent is set up, how it performs on latency, tokens and cost, what was found and how it was debugged, with traces or screenshots in the write-up.

Judging weights writing most heavily. DEV lists writing quality as the most heavily weighted criterion, asking whether the post is clear and engaging and whether it explains what was built, who it is for and why open innovation matters. Further criteria are relevance to the prompt and theme, creativity, technical execution, and optional use of partner technology where a partner category is entered. The organisers state plainly that a great project with a thin write-up will not win, and that entrants should show their work, optionally saving an agent session with DevRelay and embedding or linking it in the post.

The organisers also flag a strategic asymmetry: partner category winners come from a smaller pool of submissions, so the odds there are meaningfully higher. A single project can enter every category it genuinely uses, but the once-per-challenge win limit applies. This is the organiser's own characterisation of its prize structure, not an independently measured statistic.

Supporting resources are named but not fully specified in the post. Everything is said to live on the HF26 DEV Challenge Hub, including the schedule, every partner category and the FAQ. Each challenge page carries its exact deadline in every timezone, and DEV directs participants to follow the #hf26challenge tag for theme reveals. Free credits and promo codes are described as coming from several partners, including Tinker, Render, Backboard and ElevenLabs, claimable at hacktoberfest.com/my. Questions are directed to the MLH Discord and the post's comments.

For freelancers, designers and developers, the practical read is that this is a portfolio-building exercise with a deadline structure rather than a contribution-credit exercise. The five-week cadence and one-submission-per-challenge cap reward scoping a small, finishable project over a large one, and the requirement to write up the work means the deliverable is a public case study as much as a repository. The partner categories are the clearest lever: choosing a stack that maps to a named category, and documenting that use concretely, is the difference between competing in the general pool and a narrower one.

The tradeoffs are worth stating. Building five separate projects in one month is a substantial time commitment, and the once-per-challenge win limit means breadth does not compound into multiple wins in the same week. The open-source AI requirement is broad enough to accommodate very different stacks, which cuts both ways: it lowers the barrier to entry but makes it harder to predict what judges will reward. The emphasis on writing quality means entrants who are strong engineers but weak writers are disadvantaged relative to the stated criteria.

What remains unknown is considerable. The post does not give exact launch dates or deadlines beyond the weekly Monday cadence and the weekend opener, does not state the total number of prizes or the size of the submission pool, and does not commit to the partner category list remaining fixed across all five challenges. The value of the free credits, the terms attached to them and the availability of the Arduino UNO Q are not specified. Whether the shift away from pull requests changes participation numbers cannot be assessed from the announcement alone.

The conclusion for anyone considering entry is to treat the announcement as a scheduling and scoping document. The concrete, checkable elements are the five-challenge cadence, the shared prompt with per-challenge secret themes, the one-submission-per-challenge cap, the new-project requirement, the named prize tiers and the writing-weighted judging criteria. Everything else, including the final partner list and the exact deadlines, has to be confirmed on the challenge hub and individual challenge pages as each week opens.