Short answer: Together AI's Startup Accelerator pays up to $15K, $30K or $50K in credits depending on how much you have raised, and Fireworks AI runs a startup program that requires more than $500k in venture funding, but publishes no credit amount. Both aim at AI-native companies building on open models, both exclude reserved GPU capacity, and neither is a self-serve free tier. If you are bootstrapped, skip to the last section.
I checked both programs on their own pages on 17 October 2026. The differences matter, because one gives you a number to plan around and the other makes you apply to find out.
Side by side
| Together AI Startup Accelerator | Fireworks AI for Startups | |
|---|---|---|
| Credit amount | Up to $15K, $30K or $50K by tier | Not published |
| Funding requirement | Tiered: up to $5M, $5M to $10M, over $10M | Venture-backed, more than $500k raised, no Series D or later |
| Age of company | Not stated | Founded within the last 5 years |
| Other requirements | Application with company, funding and use-case details | Registered, privately held, for-profit, functional website |
| Credits cover | Serverless inference, dedicated endpoints, fine-tuning, instant clusters | Text and vision inference, image generation, speech to text, embeddings, fine-tuning, on-demand GPU deployments |
| Credits exclude | Reserved GPU Clusters | Reserved compute or GPU commitments, SLAs, professional services, premium support |
| Expiry | Not stated on the page | Credits expire after 1 year |
| Decision | Team reviews and gets back to you, no timeline | Team reaches out, no review time stated |
Together AI Startup Accelerator
Together AI builds a cloud for open source models: serverless inference, fine-tuning, dedicated endpoints and GPU clusters. The accelerator is the program for AI-native startups building on those models, and it is tiered by funding raised:
| Tier | Funding raised | Platform credits | Forward-deployed engineering |
|---|---|---|---|
| Build | Up to $5M | Up to $15K | 3 hours |
| Scale | $5M to $10M | Up to $30K | 6 hours |
| Grow | Over $10M | Up to $50K | 10 hours |
Notice the wording: up to. The amount is a ceiling, and the page does not say how Together decides where in the range you land. Plan around the lower number.
Beyond credits, every tier gets exclusive content on AI engineering practices such as benchmarking and evaluations, joint go-to-market support that reaches Together's 800K+ developers, community events, and introductions to a network of VCs. That last bit is worth taking seriously if you are fundraising. Credits are a one-time saving, and an introduction can be worth more.
What is not covered: Reserved GPU Clusters. If your plan is to rent a big dedicated cluster, these credits will not pay for it. They are designed for the serverless and fine-tuning workloads early companies actually run.
The application page says only to share all the details so the team can review and get back to you. It does not list required fields, selection criteria beyond the funding tiers, or timelines. The program is selection-based, so applying is not a guarantee. Our Together AI listing has the application steps.
The separate research credits
Together also has a research credits program, described as invite only and aimed at students running research projects outside formal classes. The page frames awards as a few hundred dollars and does not publish an award size, deadline or review time. It is a nice-to-have, not a plan. See the research credits listing.
Fireworks AI for Startups
Fireworks hosts fast inference for open models plus fine-tuning and on-demand GPU deployments. Its startup program is stricter on eligibility and quieter on money. All of these are required:
- A registered company with a functional website
- Privately held and for-profit
- Founded within the last 5 years
- Venture-backed with more than $500k raised
- Not raised Series D or later
The page does not state a credit amount, so any blog quoting a figure is either guessing or reading a perk marketplace. Third-party aggregators put it somewhere between about $5K and $10K, but that is not Fireworks' own number, and the program page says nothing of the kind. Treat it as unknown until the team replies.
What the page does say: eligible startups get credits that expire after 1 year, applied automatically to usage charges. They cover text and vision inference, image generation, speech to text, embeddings, fine-tuning and on-demand GPU deployments. They only apply to on-demand usage, so reserved compute or GPU commitments, SLAs, Fireworks professional services and premium support are excluded. When credits run out, usage moves to pay-as-you-go if a payment method is on file. You also get access to applied AI engineers, startup meetups, workshops and hackathons, higher rate limits, product roadmap sessions, cookbooks and guides, and joint marketing and case studies.
How to apply: fill in the form on the Fireworks for Startups page, then wait for the team to reach out. Eligibility questions go to startups@fireworks.ai. Our Fireworks listing has the checklist.
The self-serve alternative
If you do not meet the bar, new Fireworks accounts get a small signup credit (our catalog last recorded $1) and event promo codes sometimes add more. It is enough to benchmark a model, not to run a product. The Fireworks signup listing tracks it.
Which should you apply to?
Apply to Together if you have raised under $5M and want a number to plan around, or you want the fine-tuning and dedicated endpoint credits. The Build tier has no minimum funding stated on the page, which makes it the more open door for a pre-seed team. That is an inference from the page, which lists only an upper bound, so confirm in the application.
Apply to Fireworks if you are already past a $500k round, you are under five years old, and you want the speech and image endpoints alongside text. The 1 year expiry is the thing to plan around: do not apply until you have workloads ready to spend credits on.
Apply to both if you qualify. Nothing on either page prohibits holding both, though read the terms you are accepted under. The practical benefit is price shopping: run the same open model on both and keep the cheaper one. Fireworks AI vs Together AI pricing compares the per-token rates, and DeepInfra vs Together shows the other cheap alternative.
Honest limits of these programs
- Credits are not cash. They are priced at the provider's list rate, and open model inference is already cheap. $15K buys a lot of tokens at those prices, but check it against a cheaper host in cheapest inference 2026.
- Exclusions are the fine print. Both programs carve out reserved capacity. If your workload needs guaranteed throughput, you will pay for that part.
- Funding gates are real. Fireworks wants venture backing, and Together tiers by funding raised. Neither is a fit for a founder with a side project.
- Timelines are unknown. Neither page promises a response time. Apply before you are low on credits elsewhere.
If you are bootstrapped
You are not locked out. You can run open models on a free tier and apply to these programs once you have a round. Start with every free LLM API key you can get and how to use AI for free. For hosted GPU time, free GPU credits for startups lists what is available at any funding stage. Another route in the same family is the Baseten startup program.
Every program here is in the Perkstack catalog with the date we last checked its page, so a program that closes or changes its terms stops being recommended. Create an account to see the full application steps for each.
rest of this guide
The rest of this guide picks up at "Fireworks AI for Startups".
- Fireworks AI for Startups
- Which should you apply to?
- Honest limits of these programs
Free account. No card. 350+ verified perks.
Not ready for an account? Find out what you can claim first.
Already have an account? Log in