Cloudflare for Startups currently offers eligible technology startups $10,000, $100,000 or up to $350,000 in credits at a $0 acquisition cost, with benefits lasting one year or until consumed. Credits apply to eligible usage-based Cloudflare services, while R2 and Workers AI have product-specific caps and AI Gateway is temporarily excluded; eligibility depends on company age, funding stage, recent funding, public presence and first-time participation.
Modal for Academics offers accepted graduate students, labs, and researchers up to $10,000 in free Modal compute credits. Credits can be used for research workloads on Modal, including access to B200 and H100 GPUs, with usage metered as compute is consumed.
Runpod lists NVIDIA RTX A6000 48GB instances on Community Cloud at $0.33 per GPU-hour, while Secure Cloud is listed at $0.53 per GPU-hour on the same official model page. Runpod states that GPU Pods are billed by the second and can be terminated without a long-term contract.
Modal Starter has a $0 monthly plan fee plus metered usage and includes $30 per month of compute credits, 3 workspace seats, 100 containers and 10 GPU concurrency. Included compute covers Functions, Sandboxes and Notebooks, but Shared Endpoints are billed per token from the first request; region and execution settings can increase rates.
AutomationDocker
Risk: Free credits offset usage but do not make all workloads free; region selection and non-preemptible execution can materially increase compute cost. Credits are recurring plan allowance, not cash and not a fixed GPU-hour entitlement.
AWS Free Tier gives new customers $100 in AWS credits at account creation and up to an additional $100 for completing qualifying activities, for up to $200 total. The Free account plan lasts up to six months or until credits are exhausted, and includes access to more than 30 services with Always Free monthly usage limits.
Runpod lists NVIDIA RTX A5000 24GB instances on Community Cloud at $0.16 per GPU-hour and Secure Cloud at $0.27 per GPU-hour. Runpod bills Pods per second with no minimum duration; Community Cloud is lower-cost capacity on third-party hardware, and Runpod states there are no ingress or egress charges.
General
Risk: Community Cloud uses third-party hardware, so host quality and availability can vary. 24GB VRAM limits larger-model workloads compared with 48GB+ GPUs.
Cloudflare Workers AI includes a free allocation of 10,000 Neurons per day on Workers Free and Paid plans, resetting daily at 00:00 UTC. Paid usage above the free allocation is $0.011 per 1,000 Neurons, while some frontier models require a paid billing method.
Runpod Serverless Flex lists its 16GB GPU class at $0.58 per GPU-hour, billed per second. The tier includes GPUs such as A4000, A4500, RTX 4000 and RTX 2000 and scales to zero for bursty inference workloads.
Risk: GPU model is class-based rather than guaranteed; workload performance can vary across hardware within the same tier.
Bump Models links current opportunities back to named provider sources. Provider claims are evidence inputs; they do not control Bump Models rankings or scores.
Gemini 3.8 Live publishes audio input/output minute rates, while OpenAI splits voice pricing across GPT-Live-1, realtime models, transcription and translation. Compare the billing layers before estimating agent cost.
Beam says new accounts include $30 in credit refreshed monthly, while its live onboarding asks users to add credits before workloads run. The useful rule is to verify the account balance before treating the benefit as spendable.
Where is the value?
A useful deal is more than a low number: timing, quality and restrictions all matter.
Current offer value
Deal Score describes how attractive the current offer is now.
Intrinsic compute quality
Compute Score stays independent from the current deal and is never fabricated when unavailable.
Verification recency
Checked time tells you when the underlying offer was most recently re-verified.
How value works
1. Read the native price
Provider-listed currency and billing context remain the primary factual price.
2. Separate Deal and Compute scores
One measures the current offer; the other measures intrinsic resource quality.
3. Check recency and evidence
Act only with a clear checked time, source evidence and visible restrictions.
Why trust the data?
The frontend fails closed rather than filling missing facts with invented values.
Evidence visible
Source and freshness remain inspectable on public resource surfaces.
Native price first
Normalized values are analytical aids, not universal tax-inclusive prices.
Commercially isolated
Affiliate or sponsored state does not alter rankings, scores or publication decisions.