Together AI Startup Accelerator: Up to $50K in Platform Credits

The Together AI Startup Accelerator gives selected AI startups up to $15K, $30K or $50K in Together AI platform credits, depending on how much money they have raised, plus a few hours of hands-on help from Together AI engineers. It is not automatic: you apply, and Together AI picks who gets in. The credits work on serverless inference, dedicated endpoints, fine-tuning and instant clusters, but not on Reserved GPU Clusters.

Together AI has no free trial for ordinary sign-ups. Its billing docs say a new organization's first credit purchase must be at least $15. For most startups, this accelerator is the main way to get free Together AI usage. See Together AI signup credits (ended) for the self-serve rules.

What you get, tier by tier

Together AI puts startups into one of three tiers based on total funds raised. The tier sets the maximum credit amount and the number of forward-deployed engineering hours. These figures come from the official program page and the October 15, 2025 launch post, which match each other.

Each amount is a ceiling ("up to"). Together AI does not explain how it sets the exact figure inside a tier, so do not plan your runway around the maximum until you have an acceptance and can see the balance in your account.

Benefits beyond credits

• Forward-deployed engineering: free time with Together AI engineers to tune your architecture and work through performance bottlenecks.
• Exclusive content: best practices on AI engineering, plus benchmarking and evaluation resources.
• Joint go-to-market: marketing, co-selling and other promotion to Together AI's developer base. The program page puts that audience at 800K+ developers; the launch post said over 1 million.
• Community: invitations to exclusive events with other founders and AI researchers. The launch post names Together AI chief scientist Tri Dao, co-creator of FlashAttention, among the research leaders members can meet.
• VC network: introductions to venture investors and their portfolio ecosystems.

The launch post named Corridor (an AI code security company) and PlayerZero as members of the first cohort, and quoted Kleiner Perkins partner Mamoon Hamid. Neither the post nor the program page says Kleiner Perkins is a formal program partner.

What the credits cover, and what they don't

Accelerator credits work across most of Together AI's paid products. Here is how they map onto the platform, using prices from Together AI's public pricing page at the time of writing. Prices change often, so check the pricing page before budgeting.

Leaving out Reserved GPU Clusters matters most to teams planning long pre-training runs. The credits suit inference-heavy products and short bursts of training or fine-tuning, not committed reserved capacity.

How far the credits go: worked examples

These examples use the listed prices above. They show orders of magnitude, not quotes, and they ignore cached-input discounts and price changes.

• Build tier, cheap open model: with gpt-oss-120B at $0.15 per million input tokens and $0.60 per million output tokens, a workload of 3 million input tokens and 1 million output tokens costs (3 × $0.15) + (1 × $0.60) = $1.05. A $15,000 grant would cover about 14,000 such blocks, or roughly 42 billion input and 14 billion output tokens in that ratio.
• Larger dense model: Llama 3.3 70B at $1.04 per million tokens in both directions costs $1.04 per million total tokens, so $15,000 covers about 14.4 billion tokens.
• Dedicated endpoint: one H100 dedicated endpoint at $5.49/hour costs $5.49 × 24 = $131.76 per day, or about $3,953 for 30 days. A $30,000 Scale grant would run one such endpoint around the clock for roughly seven and a half months, or two endpoints for under four months.
• Instant cluster burst: eight H100s at $3.99 per GPU-hour cost 8 × $3.99 = $31.92 per hour. A 48-hour fine-tuning or experiment run comes to $1,532.16.

Together AI is prepaid. Its billing docs say that if your balance reaches zero, API access is suspended until you add credits. Creating a dedicated endpoint, fine-tuning job or evaluation job also requires enough balance to cover its cost. Watch your balance as the grant runs down so production traffic is not cut off.

Who should apply

Together AI aims the program at startups building AI-native applications: products where model inference or training is central, not an add-on. The only published criterion is funds raised, which sets your tier. Together AI does not publish rules on company age, country, revenue, team size, or whether you must be new to the platform.

• Good fit: teams serving open-weight models (Llama, Qwen, DeepSeek, gpt-oss and others) in production, teams that want to fine-tune open models, and teams that need short-term GPU clusters for experiments.
• Weaker fit: companies that mainly call closed models such as Claude or GPT through their makers' APIs. Those teams may get more from Claude for Startups or OpenAI for Startups.
• Unclear: bootstrapped companies with no outside funding. The lowest tier covers "up to $5M" raised, and nothing says zero is excluded, but acceptance is selective.

How to apply, step by step

1. Create a Together AI account at api.together.ai if you don't have one, so you have an organization ready for the credits. Signing up costs nothing, though using the API normally needs a $15 minimum first purchase.
2. Open the program page at together.ai/startup-accelerator and scroll to the application form. The page asks you to share your details so the team can review them and get back to you.
3. Describe your product and funding accurately. Your total funds raised decides whether you are considered for Build, Scale or Grow, so give the figure you can back up.
4. Explain your workload. Together AI doesn't publish its scoring, but the program is aimed at AI-native products, so a short, concrete description of which models you run and roughly how much traffic you expect helps a reviewer see the fit.
5. Wait for the team's response. Together AI does not publish a review timeline.
6. If accepted, confirm the credit balance in your account and ask how to book your forward-deployed engineering hours. The program page does not say how those hours are scheduled or whether they expire.

Getting the most from the grant

• Start serverless, move to dedicated later. Per-token serverless pricing costs nothing when idle, while a dedicated endpoint bills every hour it runs. Use serverless until steady traffic makes dedicated capacity cheaper per request.
• Benchmark cheaper models early. Prices on Together AI range widely, from cents per million tokens for small models to several dollars for large ones. Evaluating a smaller or mixture-of-experts model against your quality bar can stretch the same grant many times further.
• Use the engineering hours on a specific problem. Three to ten hours goes fast. Bring a concrete latency, throughput or cost issue rather than a general introduction.
• Shut down idle endpoints and clusters. Hourly resources keep billing until you stop them.
• Plan for the end of the credits. Add a payment method and consider auto-recharge, which Together AI's billing docs say works only with a card as the default payment method, before the grant runs out.

How it compares with other startup programs

Together AI's grant is smaller than the biggest cloud programs but is spent directly on hosted open-model inference and fine-tuning, with no infrastructure to manage. Many startups stack several programs.

If you only need to prototype, you may not need credits at all. Several providers offer free API tiers. See our roundup of free AI API keys and the broader guide on how to get free AI credits.

History of the offer

• October 15, 2025: Together AI announced the Startup Accelerator with the Build, Scale and Grow tiers, the credit ceilings above and the Reserved GPU Cluster exclusion.
• As of October 2026: the program page still lists the same three tiers, amounts and engineering hours. The joint go-to-market audience now reads 800K+ developers on the program page, compared with "over 1 million" in the launch post.
• Separately: Together AI's billing docs say it does not currently offer free trials for regular sign-ups, which makes the accelerator its main free-credit route for companies.

What is not published

Together AI does not publicly state how long applications take to review, when accelerator credits expire, whether you must be a new customer, how forward-deployed engineering hours are scheduled, or any country restrictions. Ask about these in your application or acceptance conversation, and get the answers in writing if they affect your plans.