Best offer on record: Ollama local inference runtime (free, open source). 3 verified findings, checked 2026-10-22.
Offers
Ollama local inference runtime (free, open source) high confidence
Download Ollama's desktop app or CLI and run open-weight models (Llama, Qwen, Mistral, Gemma, gpt-oss, DeepSeek, etc.) on your own CPU/GPU at $0. No account, no API key, no per-token billing; local usage is unlimited and it works fully offline / air-gapped. Permissive open source: the repo LICENSE file is MIT (the 'known so far' note said Apache-2.0 — that is outdated). Scale per ollama.com homepage: 9M+ monthly installs, 1B+ model downloads, 200T+ tokens served. This is the core 'free compute' angle for students/startups: you pay for hardware, not for the runtime.
Eligibility
Anyone, worldwide — no .edu email, no startup age/funding cap, no credit card, no region limits. Identical for students and startups as for everyone else.
How to claim
Download from ollama.com (desktop app or CLI), install, then `ollama run <model>` (e.g. `ollama run llama3`). Models pull from ollama.com's registry. No signup required for local use.
Expiry
None — permanent, no time limit or renewal, for as long as Ollama is maintained.
Gotchas
You pay for your own hardware (RAM/VRAM + electricity + depreciation) — that is the real cost surface; no free GPU or free training compute is bundled. Ollama is an inference/serving runtime, not a notebook/training platform; fine-tuning needs separate tooling. License is MIT, not Apache-2.0. No egress/egress-style fees locally (data stays on your machine).
Ollama Cloud Free tier ($0 hosted GPU inference) high confidence
Ollama Cloud's Free plan is $0/mo and includes a small 'Starter usage credits' allowance (a modest monthly usage pool for a set of 'starter models'; no dollar figure is published). 1 concurrent request. Zero data retention, hosted in the US/Europe (some Qwen models route to Singapore); prompts are never logged or trained on. No service fees, no subscription. Model pay-as-you-go rates range ~$0.015/M cached-input (nemotron-3-super) up to $15/M output (kimi-k3); off-peak half-pricing applies on some models outside 12:00–18:00 UTC weekdays and all weekend. Free tier structure (monthly starter credits, no 5-hr/weekly caps) effective Aug 31, 2026.
Eligibility
Anyone who creates a free account at ollama.com — no .edu, no startup age/funding cap, no region gate. Credit card is NOT required for the free tier; a payment method is only needed if you opt in to add pay-as-you-go credits to unlock all models.
How to claim
Create a free Ollama account at ollama.com → in the CLI run `ollama signin` → run a cloud model (e.g. `ollama run qwen3-coder:480b-cloud`, `gpt-oss:120b-cloud`) or call Ollama's OpenAI-compatible API. Starter usage is included; add credits to go beyond it.
Expiry
Free tier has no expiry; starter credits refresh monthly. No time limit on the $0 tier.
Gotchas
Starter-credit amount is not published (no dollar figure). Concurrency capped at 1; requests beyond are queued (a full queue rejects new requests). Overage = pay-as-you-go at published per-token rates (real cost). It is inference only — no free notebooks, no training GPU. This is a permanent free tier, not a trial/credit drop.
No Ollama startup program, student/education discount, or nonprofit offer high confidence
As of 2026-10-22 Ollama offers NO dedicated startup program, NO student/academic discount, and NO nonprofit tier. ollama.com/pricing lists only Free / Pro / Max / Team / Enterprise — no discounted or special student/startup/edu/nonprofit plan. Discount aggregators (costbench, aitoolsatlas) print generic 'startup/education/nonprofit program' templates but explicitly state no Ollama-specific terms are verified. A third-party 2026-09-04 review confirms: 'Ollama has no student or academic discount' — the pitch is that the core software is already free and unrestricted, so there is nothing to discount.
Eligibility
N/A — none of these programs exist. The only free offers available to students/startups are the general ones above (local runtime + Cloud Free tier), which need no special eligibility.
How to claim
Nothing to claim; use the general Free tier / local runtime.
Expiry
N/A
Gotchas
Treat auto-generated 'Ollama student discount / startup program / nonprofit discount' pages (costbench, aitoolsatlas, nozz's paid Founding-Member upsell, trytrick) as unverified, not official Ollama programs; confirm anything on ollama.com before relying on it.
LICENSE CORRECTION: the 'known so far' note said Ollama is Apache-2.0 — the repo's LICENSE file is actually MIT (permissive open source; free and unrestricted). MATERIAL PRICING CHANGE: Ollama Cloud pricing was overhauled on Aug 31, 2026 to transparent monthly usage-credit pools. Legacy mechanics — 5-hour session limits + 7-day weekly caps, and the 'Pro = 50x Free / Max = 5x Pro' framing — are gone; new structure: Free (starter credits, 1 concurrent) / Pro $20/mo ($200/yr ≈ $16.67/mo, $60/mo credits, 3 concurrent) / Max $100/mo ($300/mo credits, 10 concurrent, early access to newest models) / Team $500/mo (early access, $1,000/mo shared credits, unlimited users) / Enterprise (custom, volume pricing). Cloud (hosted inference) models first launched in preview Sep 19, 2025; Team plan opened Aug 31, 2026. NOTHING FREE WAS SUNSET: the free local runtime and the Cloud Free tier remain $0 — only the free tier's usage-limit mechanism changed (session/weekly → monthly starter credits). ADJACENT / THIRD-PARTY 'free GPU + Ollama' offers that appear in search are NOT Ollama's own programs: UCloud CompShare (10 hours free RTX40-series GPU after signup + ID verification; 5% discount + compute subsidy for academic/enterprise certification; hosts Ollama images) and Brown CCV 'Oscar' HPC (800+ NVIDIA GPUs, Ollama preinstalled as a module, 100+ shared open models, free to Brown accounts). Ollama provides no free notebooks or free ML-training compute — its cloud is inference-only and the local runtime is for inference/serving; training/fine-tuning requires separate tooling.