Events
- Spoke at Yotta 2026
%%
Lossless Twenty CRM: Yotta 2026 (synced 2026-10-04)
AI inference cloud (Palo Alto, founded 2022) deploying 100+ open-source models via APIs on its own inference-optimized infrastructure across US data centers. Founders Nikola Borisov (CEO), Georgios Papoutsis, Yessenzhar Kanapin. $107M Series B (2026) after $18M Series A (2025, Felicis co-led).
- Yotta touchpoint: Speaker company, The Token Wars
- Contacts: Nikola Borisov (Co-Founder & CEO)
- Event Map segment: Compute, Chips & Inference Clouds
Value Proposition & Features
DeepInfra provides production-ready AI inference infrastructure and a managed inference cloud for high-throughput workloads. It is positioned around cost efficiency, low latency, scalability, and control for organizations deploying open-source and proprietary models.
[E1]
- Dedicated GPU infrastructure: DeepInfra owns and operates its GPU infrastructure rather than merely routing requests to third parties. [E1] [6]
- Multiple service tiers: Model pricing can include Standard, Priority, and Flex tiers with different cost and performance trade-offs. [2]
- Developer APIs: Provides model-specific API endpoints, demos, and pricing information. [8]
Screenshots
No three official screenshots were identified in the returned search results.
Product Roadmap / Announcements
As of October 4, 2026, the returned results show recent model-availability announcements rather than a conventional public roadmap:
- October 3, 2026: DeepInfra announced the availability of GLM-5.3 through its API, with promotional pricing. [4]
- September 28, 2026: DeepInfra announced that GLM-5.3-Flash was available, with Standard pricing of $0.15 per 1 million input tokens and $0.50 per 1 million output tokens before promotion. [11]
- September 28, 2026: DeepInfra announced DeepSeek-V4.1-Flash, offering Standard, Priority, and Flex pricing tiers. [2]
- September 24, 2026: DeepInfra published guidance on model deprecation and listed supported model context lengths and prices. [12]
Recent Developments
- Model expansion: DeepInfra added GLM-5.3, GLM-5.3-Flash, and DeepSeek-V4.1-Flash to its hosted model offering in late September and early October 2026. [2] [4] [11]
- Pricing tiers: DeepSeek-V4.1-Flash is offered at Standard, Priority, and Flex rates, with Flex priced below Standard and Priority priced above it. [13]
- Provider comparison: DeepInfra’s own benchmark content claims the lowest blended price among 14 evaluated providers for GLM-5.3, but this is a vendor-produced comparison rather than independent validation. [6]
History and Origin Story
DeepInfra was founded in 2022 and is headquartered in Palo Alto, California.
[E1]
The available material describes its evolution as an AI inference infrastructure provider focused on operating GPU capacity and hosting open-source models, but the returned search results did not identify founders or provide a reliable, detailed founding narrative.
Fundraising History
| Round | Date | Amount | Lead investor |
| Seed and other early rounds | 2023–2025 | Not individually disclosed | No reliable source found |
| Series A | 2025 | Not individually disclosed | No reliable source found |
| Series B | May 1, 2026 | $107 million | No reliable source found |
| Total | 2023–2026 | $153.64 million reported | — |
The reported total includes the Series B, Series A, seed rounds, and other ventures.
[E1]
No reliable search result identified a complete investor list or confirmed the lead investor for each round.
Notable Team Members
No reliable source found identifying DeepInfra’s founders or notable leadership team members.
Market Sizing
Category, Market Size, and Category Growth
DeepInfra operates in the AI inference infrastructure and model-serving cloud category, particularly managed inference for open-weight and proprietary models.
[E1]
[6]
No reliable analyst-firm or financial-journalism estimate for the specific addressable market of DeepInfra’s category was found in the returned results.
Pricing
| Pricing component | Published example |
| Input tokens | $2.00 per 1 million tokens |
| Output tokens | $6.00 per 1 million tokens |
| Cached tokens | $0.20 per 1 million tokens |
DeepInfra uses token-based pricing, although rates vary by model and service tier.
[E1]
For DeepSeek-V4.1-Flash, the listed Standard rates are $0.20 per 1 million input tokens, $0.60 per 1 million output tokens, and $0.006 per 1 million cached input tokens; Priority is 1.5× Standard and Flex is 0.8× Standard. [13]
Revenue Trajectory Estimates
The available metadata reports annual revenue of approximately $3.8 million , but the returned search results do not independently explain the figure or establish its reporting date.
[E1]
Competitive Landscape
Who it's for, who it's not for
DeepInfra is aimed at teams that need managed, production-scale inference for open-source models without operating the full GPU-serving stack themselves. Its stated fit includes high-throughput applications, agentic systems, and workloads where latency, model choice, and token economics matter.
[E1]
It is less suited to users seeking only model training infrastructure, fully self-managed deployment, or a single closed-model vendor with no need for open-model choice. This is an inference from DeepInfra’s stated emphasis on hosted inference and its open-source model catalog.
[E1]
Viable Alternatives
- Baseten: A managed model-serving alternative for deploying production inference workloads; DeepInfra’s own comparison names Baseten as a benchmarked provider. [6]
- Amazon Bedrock: A broader managed foundation-model platform that may suit organizations prioritizing cloud-provider integration over a specialized open-model inference service.
- Google Vertex AI : A general-purpose AI platform with managed model access and deployment capabilities.
- Fireworks AI : A model-inference provider serving open-weight models and competing on latency, model availability, and price.
- Self-hosted GPU infrastructure: Organizations can deploy models directly on their own cloud or dedicated GPU clusters when infrastructure control outweighs operational simplicity.
DeepInfra can be evaluated alongside Requesty when comparing multi-provider access and model-routing workflows.
Competitor Table
| Competitor | Description |
| Baseten | Managed infrastructure for deploying and serving machine-learning models; cited by DeepInfra in a provider-price comparison. [6] |
| Amazon Bedrock | Cloud-provider model platform for accessing and integrating foundation models. |
| Google Vertex AI | Google Cloud’s broader platform for model development, deployment, and managed AI services. |
| Fireworks AI | Managed inference platform for open-weight and foundation models. |
| Self-hosted GPU deployment | Direct operation of model-serving infrastructure for teams requiring maximum deployment control. |
Sources
[1]: GLM-5.3 Provider Pricing Guide: Costs Compared
[2]: DeepSeek-V4.1-Flash Is Now on DeepInfra
[3]: deepinfra.com · blog · glm-5-3-flash-pricingGLM-5.3-Flash Pricing, Providers & Cost - deepinfra.com
[4]: GLM-5.3 Is Now Available on DeepInfra
[5]: Best AI Inference Platforms for Speed & Cost in 2026
[6]: Deepinfra: The Overall...
[7]: DeepSeek V4.1 Flash API - Voice - DeepInfra
[8]: DeepSeek V4.1 Flash API - Demo
[9]: DeepSeek V4.1 Flash API: Speed, Latency & Cost
[10]: AI Model Distillation: Teacher vs. Student Models - DeepInfra
[11]: GLM-5.3-Flash API Is Now on DeepInfra
[12]: Model Deprecation: Build LLM Apps That Last
[13]: DeepSeek-V4.1-Flash: Model Overview & Integration - DeepInfra
[14]: DeepSeek-V4.1-Flash Pricing Guide for Developers - DeepInfra
[15]: Small Open-Weight Models: The Hidden Cost Advantage