Together AI Startup Accelerator
Selection-based program giving startups up to $50k in Together AI platform credits for open-model inference, fine-tuning, and dedicated endpoints.
- Free credit
- $50,000
- Card required
- No
Together AI Startup Accelerator scores 9.2/10 for free value in LLM. The main constraint: Up to $15k (Build, <$5M raised), $30k ($5–10M), or $50k (>$10M) in free platform credits, plus forward-deployed engineering hours. Credits exclude reserved GPU clusters.
Free value score
9.2 / 10Show the math
Key metrics
Models
21 free · 200+ totalLimits may be stale — last re-verified outside this section's freshness window.
| Model | Family | AA | Access |
|---|---|---|---|
| Qwen3.5-122B-A10B | Qwen | 32.3 | free |
| gpt-oss-20B | GPT | 14.9 | free |
| Llama 4 Scout | Llama | 10 | free |
| Llama 3.1 405B | Llama | 8.5 | free |
| Gemma 3 27B | Gemma | 4.8 | free |
| Cogito v1 Preview Qwen 32B | Qwen | — | free |
| Devstral Small 2505 | Mistral | — | free |
| GLM 4.5V | GLM | — | free |
| Llama 3.1 70B Instruct | Llama | — | free |
| Llama 3.2 90B Vision Instruct | Llama | — | free |
| MiniMax M1 | MiniMax | — | free |
| MiniMax M2 | MiniMax | — | free |
| Mistral Small 3.2 24B | Mistral | — | free |
| Mixtral 8x22B Instruct | Mistral | — | free |
| Nemotron 3 Super 120B A12B | Llama | — | free |
| Qwen3 30B A3B Instruct 2507 | Qwen | — | free |
| Qwen3 32B | Qwen | — | free |
| Qwen3 Coder 30B A3B Instruct | Qwen | — | free |
| Qwen3 Next 80B A3B Instruct | Qwen | — | free |
| Qwen3 VL 235B A22B Instruct | Qwen | — | free |
| R1 Distill Qwen 7B | DeepSeek | — | free |
| Qwen3.7-Max | Qwen | 46 | paid |
| DeepSeek V4 Pro | DeepSeek | 44.3 | paid |
| Kimi K2.6 | Kimi | 42.8 | paid |
| GLM-5.1 | GLM | 40.2 | paid |
| Qwen3.6-Plus | Qwen | 39.6 | paid |
| GLM-5 | GLM | 39.5 | paid |
| Kimi K2.5 | Kimi | 38.1 | paid |
| MiniMax M2.7 | MiniMax | 38.1 | paid |
| Qwen3.5-397B-A17B | Qwen | 32 | paid |
| gpt-oss-120B | GPT | 23.8 | paid |
| DeepSeek-V3.1 | DeepSeek | 21.1 | paid |
| DeepSeek-R1 | DeepSeek | 20.1 | paid |
| Llama 4 Maverick | Llama | 14.3 | paid |
| Cogito v2.1 671B | DeepSeek | — | paid |
| FLUX.2 [pro] | FLUX | — | paid |
| Llama 3.3 70B Instruct Turbo | Llama | — | paid |
| MiniMax M2.5 | MiniMax | — | paid |
| Qwen3 235B A22B FP8 Throughput | Qwen | — | paid |
| Qwen3-Coder 480B A35B Instruct | Qwen | — | paid |
Showing first 21 of a larger catalog.
What you get
Official page tiers: 'Up to $15K / $30K / $50K in free platform credits' by funding stage
Credits cover serverless inference, dedicated endpoints, fine-tuning, instant clusters
Includes free forward-deployed engineering time
Access to 100+ open-source models via OpenAI-compatible API
Gotchas
- Funded startup gate
Only startups that have raised external funding qualify; the program is explicitly for venture-backed or seed-funded companies.
- Application review required
Acceptance is selection-based via an application form; Together AI's team manually reviews and approves each applicant.
- Credits capped by funding tier ($15K–$50K)
The credit pool is finite and one-time: $15K for Build (<$5M raised), $30K for Scale ($5M–$10M), or $50K for Grow (>$10M raised) — credits do not replenish.
- Cohort program — not always open
The accelerator runs in cohorts, meaning applications may only be accepted during specific windows and not continuously throughout the year.
- Funding-stage credit tier lock
The credit amount is permanently determined by funding raised at time of application; startups cannot upgrade to a higher tier after joining.
Sources
-
"Up to $15K in free platform credits* 3 hours of free forward-deployed engineering time ... Up to $30K in free platform credits* 6 hours ... Up to $50K in free platform credits* 10 hours of free forward-deployed engineering time ... * Credits do not apply to Reserved GPU Clusters"
https://www.together.ai/startup-accelerator -
"Together AI does not currently offer free trials. Access to the Together platform requires a minimum $5 credit purchase. Together AI is fully prepaid. You need a positive credit balance to use the platform."
https://docs.together.ai/docs/billing-credits -
"Kimi K2 Thinking (1T): Released November 4, 2025, Intelligence Index: 67 ... Kimi K2 Thinking is arguably the best open model in the world by benchmark score. ... DeepSeek V3.2 (685B): Released December 1, 2025, Intelligence Index: 66"
https://www.understandingai.org/p/the-best-chinese-open-weight-models -
"DeepSeek V3.1. Most teams start with serverless inference and move to dedicated endpoints at scale. $1.20 Kimi K2.6 $4.50 Qwen3.6-Plus $0.50."
https://www.together.ai/pricing
