#14 on the AGIX Leaderboard · score 34
Prime Intellect
Decentralised training of open models (INTELLECT).
| Type | Open models & research |
|---|---|
| Founded | 2024 |
| Founders | Vincent Weisser, Johannes Hagemann |
| Headquarters | San Francisco, US |
| Website | primeintellect.ai |
Prime Intellect hiring
Prime Intellect lists 39 open roles on its Ashby job board today, and 23 of them are research, model, training or inference roles. Top locations: San Francisco (38), New York (1).
- Research Engineer - Distributed TrainingSan Francisco
- Member of Technical Staff - GPU InfrastructureSan Francisco
- Member of Technical Staff - Training PlatformSan Francisco
- Member of Technical Staff - Sandbox PlatformSan Francisco
- Applied Researcher - Agentic RLSan Francisco
- Applied Researcher - Forward-Deployed RLSan Francisco
- Applied Researcher - Continual LearningSan Francisco
- Member of Technical Staff - Prime AgentSan Francisco
- Research Engineer / Scientist - Reinforcement LearningSan Francisco
- Research Engineer - Pre-training DataSan Francisco
- Research Engineer - Data and EvaluationSan Francisco
- Research Engineer - InferenceSan Francisco
Source: Prime Intellect careers (Ashby), read 2 Nov 2026. All research roles at tracked labs.
Prime Intellect funding filings
Prime Intellect has filed 2 Form D notices with the SEC under the name we track. Form D is filed after an exempt securities sale; "amount sold" is what the company reported at the time.
| Filed | Form | Amount sold | Offering size | Source |
|---|---|---|---|---|
| 1 Jul 2026 | D | $82.5M | $85.0M | SEC filing |
| 15 Jan 2026 | D | $49.9M | $64.6M | SEC filing |
Source: SEC EDGAR full-text search for "Prime Intellect", Form D. Name matches can include unrelated filers; tell us if one is wrong.
Prime Intellect research
2 papers in OpenAlex list an author affiliated with Prime Intellect (United States) in the last 12 months. The most cited:
- Prefix Sliding for efficient test-time scaling26 Aug 2026 · 0 citations
- Prime Agent: A Self-Improving RLM Harness24 Aug 2026 · 0 citations
Source: OpenAlex (CC0), institution Prime Intellect (United States), read 2 Nov 2026.
Prime Intellect models
Latest releases
- PrimeIntellect/DeepSeek-V4.1-Flash-bf1615 Oct 2026
- PrimeIntellect/Qwen3.5-4B-untied5 Oct 2026
- PrimeIntellect/Qwen3-4B-Instruct-2507-untied5 Oct 2026
- PrimeIntellect/Qwen3-0.6B-Reverse-Text-SFT-untied5 Oct 2026
- PrimeIntellect/Qwen3-0.6B-untied5 Oct 2026
- PrimeIntellect/Qwen3.5-0.8B-Reverse-Text-RL4 Oct 2026
Most downloaded
- PrimeIntellect/Qwen3-0.6B1,268,822 downloads (30 days)
- PrimeIntellect/Qwen3-1.7B337,838 downloads (30 days)
- PrimeIntellect/Qwen3-0.6B-Reverse-Text-SFT33,718 downloads (30 days)
- PrimeIntellect/DeepSeek-V4-Flash-Base-bf1615,321 downloads (30 days)
- PrimeIntellect/DeepSeek-V4-Flash-0731-bf169,973 downloads (30 days)
Source: Hugging Face Hub, organisation PrimeIntellect, read 2 Nov 2026.
Prime Intellect: quick answers
Is Prime Intellect hiring?
Yes. Prime Intellect had 39 open roles on its job board on 2 Nov 2026, including 23 research and model roles.
Who founded Prime Intellect?
Prime Intellect was founded in 2024 by Vincent Weisser, Johannes Hagemann.
How much has Prime Intellect raised according to SEC filings?
Its most recent Form D, filed 1 Jul 2026, reported $82.5M sold. Form D figures cover one offering, not total funding.
People at Prime Intellect
Similar labs
AI cloud and open-model research (RedPajama, FlashAttention authors).
29 research roles · 7 models#18Ai2 (Allen Institute for AI)Seattle, US · 2014Non-profit releasing fully open models, data and code (OLMo, Tülu, Molmo).
247 papers · 172 models#19IBM Research (Granite)Yorktown Heights, USIBM's open Granite model family.
467 papers · 79 models#23Databricks Mosaic ResearchSan Francisco, USDatabricks' model research group (DBRX).
23 research roles#37Salesforce AI ResearchPalo Alto, USSalesforce's research group (xLAM agent models).
286 papers · 13 models#41Snowflake AI ResearchSan Mateo, USSnowflake's research team (Arctic models).
12 research roles · 5 modelsSomething wrong or missing? Send a correction. We fix sourced errors within a day.