Perplexity Decider Free for a Week, Plus Compute Grants and Credit Matching

This Week's Best AI Freebies: Perplexity's New Models Free for a Limited Time

Perplexity's newly released Decider V1.3 and Decider Flash V1.3 are currently available for free on OpenRouter. The official post says "For a week" without specifying an exact time. Based on the post date, the window likely closes around November 19 — if you want to try them, this week is the easiest opportunity.

This isn't the kind of "fake free" where you have to supply your own GPU and deploy it yourself. You can simply select Decider V1.3 in OpenRouter and use it with zero hardware barriers. If you want to host the model yourself, you can also download the weights from Hugging Face. The Flash version is smaller and can run locally.

Two Ways to Use the Free Window

Pure trial users: Go through OpenRouter — just register and select the model. Ideal for developers who want to quickly compare Decider's output quality against other models. A week is enough to run a round of small-scale evaluation.

Local deployment users: Download the Flash weights and self-host. Suitable for scenarios involving data privacy sensitivity or fine-tuning. Note that running locally means you bear the compute costs yourself — the free window only covers API calls.

Both models are on the V1.3 version line. The official post doesn't go into architecture or benchmark details, so you'll have to verify performance hands-on. For those doing model selection, this week is more of a "zero-cost trial window" than a long-term solution.

OpenRouter is a unified API platform that aggregates multiple model providers. Developers need only one account and API key to call models from different vendors through an OpenAI-compatible interface, eliminating the hassle of separate registrations, price comparisons, and SDK integrations. Billing is typically token-based, and the platform occasionally features free models offered by vendors — such free calls are usually subsidized by the model provider to attract new users and collect feedback. For those wanting quick side-by-side model comparisons, OpenRouter's value lies in its extremely low switching cost: change one model name to swap engines, making it well-suited for small-scale evaluation within a one-week window.

Hugging Face, meanwhile, is the hub for open-source models and datasets, where model weights, inference scripts, and community discussions are all concentrated. Downloading weights from Hugging Face for self-hosting means the model runs on your machine or private cloud, data stays within your internal network, and you can do quantization, fine-tuning, and other secondary processing. The trade-off is that you need to provision GPU memory, inference frameworks, and operations yourself — small versions like Flash are specifically designed for local deployment.

Open-Source AI Projects Can Apply for Up to $10,000 in Compute Grants

The second resource comes from Infirst.net: compute grants for open-source AI projects, offering up to $10,000 in free compute. It's ongoing and available anytime.

To apply, submit a project description and intended use on the grant page. The team reviews on a rolling basis with no fixed deadline. The key phrase here is "rolling review" — meaning there's no deadline pressure, but also that the review timeline is opaque. Submitting earlier theoretically gets you into the queue sooner.

What Kinds of Projects Are More Likely to Pass

The official post names three priority areas: novel applications, inference performance optimization, and developer tools. This essentially outlines the review preferences — it's not enough to just have an AI idea; it needs to fall into one of three buckets: "new application forms," "making inference more efficient," or "lowering the barrier for others to develop."

The threshold hinges on the word "open-source." You need a clearly defined project first, and after passing review, the team will contact you to coordinate — credits aren't disbursed just for submitting. In other words, this is a resource you need something substantial to redeem. Empty applications are unlikely to pass.

Rolling review is a mechanism with no unified deadline, where submissions are reviewed in batches on a first-come, first-served basis. It contrasts with centralized review on fixed deadlines: the advantage is no time pressure — submit early, queue early; the downside is that both the review timeline and pass rate are opaque, making it hard for applicants to estimate wait times. For funders, rolling review facilitates continuous project intake and controls per-batch review volume. For applicants, projects with complete materials and clear goals are more likely to be processed quickly in the queue, while repeated information supplements slow down progress.

Cognition Offers $15,000 in Credits for Startup Teams

The third resource is Cognition's Devin for Startups, also ongoing and available anytime.

Submit your team and project introduction on the Devin Startups page, and upon approval, $15,000 in Devin credits are deposited into your account that same day. The subsequent matching mechanism is worth noting: within the next $100,000 of usage, the official team will match 1:1 whatever the team actually pays, up to a matching cap of $50,000.

The design is essentially "get started first, talk later" — use the free credits to embed Devin into your workflow, and once you've validated the value and are willing to pay, the official team adds an equal amount. For teams already using AI coding agents, 1:1 matching effectively doubles your budget.

Applications are processed on a rolling basis with no fixed deadline and unlimited slots. Teams can collaborate on usage. It's recommended to prepare team and project materials before submitting to avoid delays from back-and-forth supplements.

Devin is an AI coding agent launched by Cognition. Unlike completion-style code assistants, it can autonomously break down tasks, read and write code in a sandbox environment, run commands, and submit changes — positioning it closer to a virtual engineer you can delegate tasks to. Such agents typically bill by usage; the larger the team and the more complex the tasks, the higher the consumption. Matching is a common growth tactic: whatever the user actually pays, the platform adds an equal amount of credits proportionally, lowering the cost of trial while converting free credits into real usage habits.

How to Prioritize the Three Resources

From the perspective of "zero-cost startup difficulty," the order is clear:

Perplexity Decider free window is the lightest — no project or team needed, just open OpenRouter and use it. The only constraint is the time window. Suitable for everyone to grab first.

Infirst.net compute grants are in the middle — you need a real, open-source AI project, and the reward is up to $10,000 in compute. Suitable for teams already pushing forward an open-source project.

Cognition Devin credits lean toward commercialization teams — $15,000 in credits plus 1:1 matching offers the most value for startups already evaluating or using AI coding agents.

What all three have in common is that none require you to pay upfront. The difference lies in prerequisites: one needs time, one needs a project, one needs a team. Pick whichever suits your situation best — no need to chase all three.