Key takeaways
- Top pick: Mistral AI, best for: enterprises requiring data sovereignty. Mistral says its Series D is the largest equity round ever completed by a European technology company.
- Also near the top: Moonshot AI, best for: Developers who want very large open models for coding and agents; and Together AI, best for: AI engineering teams building production-scale applications.
- Free to start: Moonshot AI, Reflection AI and Ollama.
- How it was ranked: 10 startups compared. Each verdict is below.
Best Open-Source AI Startups: Top 10 Ranked. Mistral AI ranks first, with a Lister Score of 97.3.
Mistral AI leads open-weight AI globally, with frontier models backed by Europe's largest tech funding round.
Moonshot AI's Kimi models are the strongest open-weight choice from China,
Together AI is the main cloud for running open models in production, and
Black Forest Labs' FLUX leads open image generation.
Ollama is the simplest way to run open models locally.
Best Open-Source AI Startups in 2026, compared| RANK | ENTRY | BEST FOR | SCORE | FROM |
|---|
| 1 |  Mistral AIMistral says its Series D is the largest equity round ever completed by a European technology company. | enterprises requiring data sovereignty | 97.3 | – |
|---|
|
| 2 |  Moonshot AIIn October 2026 the Japan Times reported Moonshot was eyeing a Hong Kong listing after its valuation reached about $50 billion. | Developers who want very large open models for coding and agents | 96.0 | Free (open weights) |
|---|
|
| 3 |  Together AITogether says customers include Cursor, Decagon, ElevenLabs and Suno, and that Decagon cut its inference costs sixfold. | AI engineering teams building production-scale applications | 94.8 | – |
|---|
|
| 4 |  Black Forest LabsBlack Forest Labs says Adobe, Canva, Meta and Microsoft are building on its technology. | Developers and platforms adding image generation | 93.5 | $0.048 per image (FLUX 3) |
|---|
|
| 5 |  Reflection AIReflection published Beam's weights on Hugging Face under Apache 2.0 on 16 October 2026. | Developers who want a US-built open model under a permissive licence | 92.2 | Free (open weights) |
|---|
|
| 6 |  OllamaOllama says 8.9 million developers use it, including teams at 85 percent of Fortune 500 companies. | Developers and API automation | 90.9 | Free (MIT license) |
|---|
|
| 7 |  LangChainLangChain raised a $125 million Series B in October 2025. | engineering teams building agents | 89.6 | – |
|---|
|
| 8 |  InferactInferact says vLLM has a community of more than 2,000 contributors. | AI engineering teams requiring high-performance inference | 88.3 | – |
|---|
|
| 9 |  Nous ResearchNous says Hermes Agent has been cloned more than 24 million times. | Developers who want open, user-controlled models and agents | 87.0 | Free |
|---|
|
| 10 |  Stability AIStability AI's August 2026 round brought its total fundraising to $232 million, according to TechCrunch. | Creative studios and developers building on open media models | 85.7 | Free (open weights) |
|---|
|
|
How to choose
Start from what you need it for. Where each entry fits best:
- Mistral AI (#1): enterprises requiring data sovereignty
- Moonshot AI (#2): Developers who want very large open models for coding and agents
- Together AI (#3): AI engineering teams building production-scale applications
- Black Forest Labs (#4): Developers and platforms adding image generation
- Reflection AI (#5): Developers who want a US-built open model under a permissive licence
- Ollama (#6): Developers and API automation
- LangChain (#7): engineering teams building agents
- Inferact (#8): AI engineering teams requiring high-performance inference
- Nous Research (#9): Developers who want open, user-controlled models and agents
- Stability AI (#10): Creative studios and developers building on open media models
Each entry's full verdict is below.
The ranking, in detail
01
Frontier open models you own
Mistral AI, founded in Paris in 2023, releases open-weight language models alongside commercial models, its Le Chat assistant and enterprise deployment tools. It closed a 3 billion euro Series D in September 2026, led by Samsung Electronics, at a post-money valuation above 21 billion euros.
PROS- The AI tool offers open-weight models that customers retain ownership of.
- The AI tool allows deep customisation through its Forge platform.
- Best for
- enterprises requiring data sovereignty
97.3
02
Kimi open-weight models from Beijing
Moonshot AI, founded in Beijing in 2023, builds the Kimi family of large language models and releases them as open weights for developers to download and run. It raised $2 billion at a $20 billion valuation in May 2026, according to TechCrunch.
PROS- Some of the largest open-weight models
- Fast release pace
CONS- Plans a public listing
- Running the biggest models needs heavy hardware
- Best for
- Developers who want very large open models for coding and agents
- Where it falls short
- Running its biggest models requires heavy hardware.
- Pick something else if
- you lack access to heavy hardware.
96.0
03
The open source AI stack and AI Native cloud
Together AI, founded in 2022 in San Francisco, runs a cloud for training, fine-tuning and serving open-source models. It raised an $800 million Series C in July 2026.
PROS- The AI tool provides serverless inference with no infrastructure management required.
- The AI tool enables managed fine-tuning jobs without managing training infrastructure.
- Best for
- AI engineering teams building production-scale applications
94.8
04
FLUX image generation models
Black Forest Labs, founded in 2024 in Freiburg, Germany, by researchers behind Stable Diffusion, makes the FLUX image generation models, releasing some as open weights and selling others through an API. It raised a $300 million Series B at a $3.25 billion valuation in December 2025.
PROS- Open-weight models for developers
- Used by large creative platforms
CONS- Focused on visual media only
- Prices vary by model and resolution
- Best for
- Developers and platforms adding image generation
- Where it falls short
- It only supports visual media and does not offer text or audio generation capabilities.
- Pick something else if
- you need multi-modal AI solutions beyond images.
93.5
05
Beam open-weight frontier model
Reflection AI, a Brooklyn-based company founded in 2024, builds open-weight frontier models for coding, reasoning and agents. It raised $2 billion in October 2025 and released the weights for Beam, a 501 billion parameter mixture-of-experts model, under an Apache 2.0 licence in October 2026.
PROS- Permissive open licence
- Very well funded
CONS- First model only just released
- Short track record
- Best for
- Developers who want a US-built open model under a permissive licence
- Where it falls short
- It has a short track record and its first model was only just released.
- Pick something else if
- you need a proven history.
92.2
06
The easiest way to run large language models locally
Ollama, based in Palo Alto, makes an open-source tool for downloading and running open models locally, plus cloud options. It announced $88 million in funding in July 2026, with Benchmark, Theory Ventures and 8VC among its backers.
PROS- Universal compatibility with local AI tools and frameworks
- Simple installation and CLI-first design for developers
CONS- Requires command-line familiarity for initial setup
- Best for
- Developers and API automation
- Where it falls short
- It requires command-line familiarity for initial setup.
- Pick something else if
- you need a graphical interface for setup.
90.9
07
The open agent platform to own your intelligence
LangChain, founded in 2022 in San Francisco, maintains the open-source LangChain, LangGraph and Deep Agents frameworks for building AI agents, and sells LangSmith for testing and monitoring them. It raised a $125 million Series B in October 2025.
PROS- LangChain offers a framework agnostic platform for observing and deploying agents.
- LangChain allows users to choose closed or open models for their workload.
- Best for
- engineering teams building agents
89.6
08
Frontier inference you control, optimised for your models and hardware.
Inferact, founded in San Francisco by the creators and maintainers of the open-source vLLM inference engine, funds vLLM development and is building a commercial inference product on top of it. It raised a $150 million seed round in January 2026, led by Andreessen Horowitz and Lightspeed, at an $800 million valuation.
PROS- The AI tool provides day-zero support for new models and hardware.
- The AI tool offers autoscaling to reduce unnecessary compute spend.
- Best for
- AI engineering teams requiring high-performance inference
88.3
09
Hermes models and open-source agent
Nous Research releases the Hermes family of open models and Hermes Agent, an open-source AI agent. It raised $90 million in October 2026 from investors including NVIDIA, Microsoft's M12, Samsung and Union Square Ventures.
PROS- Large open-source following
- Fully open releases
CONS- Smaller budget than frontier labs
- Limited enterprise support
- Best for
- Developers who want open, user-controlled models and agents
- Where it falls short
- Smaller budget than frontier labs.
- Pick something else if
- you need robust enterprise support.
87.0
10
Stable Diffusion open image and media models
Stability AI, based in London and founded in 2019, makes the Stable Diffusion family of open-weight image models and models for audio, video and 3D. It raised a $76 million Series B in August 2026 backed by Universal Music Group, Warner Music Group, Sony Music Group and Electronic Arts.
PROS- Well known open image models
- Backing from all three major record labels
CONS- Open release pace has slowed
- Focus shifting to licensed music tools
- Best for
- Creative studios and developers building on open media models
- Where it falls short
- Open release pace has slowed.
- Pick something else if
- you need rapid open model updates.
85.7
Put your company on this ranking
Ask us for a quote for one, two or three years. Nothing recurring.
Positions #4 to #10 are available on request. Contact us.
Not on this ranking?Tell us about your company and we will reply by email.
Frequently asked questions
What are the best Open-Source AI Startups?
Mistral AI ranks first on Lister. Mistral AI, founded in Paris in 2023, releases open-weight language models alongside commercial models, its Le Chat assistant and enterprise deployment tools. It closed a 3 billion euro Series D in September 2026, led by Samsung Electronics, at a post-money valuation above 21 billion euros.
What are the top 10 Open-Source AI Startups?
1. Mistral AI; 2. Moonshot AI; 3. Together AI; 4. Black Forest Labs; 5. Reflection AI; 6. Ollama; 7. LangChain; 8. Inferact; 9. Nous Research; 10. Stability AI.
Sources
Where we checked each company's facts, and when.
Some links on Lister are affiliate links. Rankings are not affected.
Related rankings
