Best offer on record: llama.cpp itself — free open-source local engine (no hosted offer). 3 verified findings, checked 2026-10-22.
Offers
llama.cpp itself — free open-source local engine (no hosted offer) high confidence
llama.cpp is an MIT-licensed C/C++ library for LLM/VLM inference and training. It is $0 to use with unlimited local use and commercial use; there is NO free tier, trial credits, startup program, or education license, because llama.cpp is a self-hosted library, not a cloud/compute provider. The only cost is the hardware you run it on (consumer GPU/CPU; GPU backends CUDA/ROCm/Vulkan/Metal) plus any cloud GPU you rent to host it. This is the accurate core answer to 'does llama.cpp offer free compute to students/startups?' — the software is free, but no hosted free-compute offer exists from it.
Eligibility
Anyone, no account, no .edu, no credit card, no region limit; bounded only by your own hardware. No student/startup-specific program exists for the llama.cpp project.
How to claim
Nothing to claim. Download a prebuilt binary or build from source at github.com/ggml-org/llama.cpp (or Docker). License text: github.com/ggml-org/llama.cpp/blob/master/LICENSE (MIT, 'Copyright (c) 2023-2026 The ggml authors').
Gotchas
Costs shift to hardware: 20–100GB+ model storage, GPU compute, electricity; idle always-on cloud GPUs still bill (~$400–450/mo T4 at startup scale per third-party analysis). Free software does not mean free compute. No training-notebook front is hosted by llama.cpp.
Meta Llama Startup Program (adjacent — for the Llama MODEL, not llama.cpp) high confidence
Meta's program reimburses early-stage US startups for using Meta's open-weight Llama models via hosted APIs on cloud inference providers: up to $6,000 USD/month for up to 6 months = up to $36,000 total, plus hands-on technical support from the Llama team. This is a REIMBURSEMENT of actual Llama API spend, not a grant, and it funds the Llama model ecosystem — it is distinct from the llama.cpp library.
Eligibility
Early-stage startups incorporated in the United States; raised less than $10M USD; at least one developer on staff. Target industries: tech/software, financial services, healthcare/life sciences, telecom, retail/eCommerce. US-only (no EU/Canada). No .edu required (not a student offer).
How to claim
Apply via the Meta Llama Startup Program page (go.fb.me / lnkd.in links from the 2025-05-21 announcement). Selected members receive monthly reimbursements up to the $6K cap over up to 6 months.
Expiry
Initial-cohort applications closed 2025-05-30 (6:00pm PT); first cohort announced 2025-06-17 (~1,000+ applications). Per-member reimbursement window is up to 6 months. As of 2026-10-22 there is NO official confirmation the program has reopened; a third-party aggregator (grantedai.com) lists it 'Active / check source'.
Gotchas
Reimbursement only for hosted Llama API costs (not on-prem/llama.cpp self-host, not training infra) — so it does not offset llama.cpp local runs. Monthly cap is 'up to' $6K, not guaranteed. US-only and <$10M funded; no new-cohort confirmation found for late 2026. Feedback on Llama is implicitly expected in exchange.
AWS + Meta 'Building with Llama' program (adjacent — Llama model on AWS) medium confidence
A 6-month virtual program co-run by AWS and Meta for ~30 promising US startups building with Llama. Offer: up to $200,000 in AWS promotional credits, plus office hours with Meta/AWS engineers, dedicated Meta Discord, AWS Solution Architect access, and model customization/inference guidance. Distinct from the smaller Meta Llama Startup Program and from llama.cpp.
Eligibility
US-based startups that have raised money, less than 10 years old, available for a 6-month virtual program. Selected startups must enroll in AWS Activate to draw the credits. No .edu (startup, not student, offer).
How to claim
Apply through the AWS Startups 'Building with Llama' landing page; selected startups sign up for AWS Activate and receive up to $200K promotional credits over the 6-month build window.
Expiry
Announced 2025-07-21; applications due 2025-08-08, AWS notified selectees by 2025-08-29; a single cohort of ~30 US startups. Landing page still live but this was a cohort-specific program; no confirmed 2026 re-open.
Gotchas
Credits are AWS promotional (consumable against AWS usage, not cash), require AWS Activate enrollment, and are gated by selection (30 spots). The direct AWS page could not be re-fetched today (network block); terms verified via official page snippet + Meta blog. Credits expire per AWS promotional-credit terms; does not fund self-hosted llama.cpp compute.
llama.cpp (the named provider) has NOT sunset and is not a compute vendor: it remains free, MIT-licensed, with unlimited local use; there is no free tier, trial credits, student/education offer, or startup program to sunset — those categories simply do not apply to a self-hosted library. The 'free-compute' offers people associate with the 'Llama' name in 2025-2026 belong to Meta's Llama model and its cloud partners, NOT to llama.cpp: (1) Meta Llama Startup Program ($6,000/mo up to 6 months = $36K reimbursement) — initial cohort closed 2025-05-30; no official confirmation it reopened as of 2026-10-22 (third-party aggregators list it 'Active, check source'). (2) AWS + Meta 'Building with Llama' (up to $200K AWS credits, 6-month, ~30 US startups) — cohort applications due 2025-08-08, a cohort-specific offering. Both are startup (US, funded/<$10M or <10 yrs old) programs tied to the Llama model ecosystem, not llama.cpp; neither is a .edu student offer. Key disambiguation to carry forward: llama.cpp = local engine; the credit programs = Meta Llama model via cloud APIs/GPU, so they do not reimburse local llama.cpp runs. AWS 'Building with Llama' page was network-blocked today, so its current enrollment status is unconfirmed.