What Tablif read on OLMo-core 3's site
OLMo-core 3 describes itself as: “Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs”
- Stage: Live, no waitlist
- Alternative to: Megatron-Core
What it is
Calls itself “training infrastructure”.
- Calls itself: training infrastructure
“Olmo-core 3 is Ai2's open training infrastructure for large mixture-of-experts language models”
Where it puts itself
Positions itself as an alternative to Megatron-Core.
- Alternative to: Megatron-Core
“NVIDIA’s Megatron-Core is an established option for training large MoEs. Olmo-core 3 brings an integrated MoE training stack”
Who it sells to
Sells to developers. Built for small businesses. Made for academic researchers, developers and researchers.
- Sells to: developers
- Company size: small businesses
- Made for: academic researchers · developers · researchers
“…researchers and developers can use Olmo-core 3 to train their own MoEs”
“…putting advanced model development out of reach for many academic researchers and smaller labs.”
How it runs
Open source. AI is the core of the product.
- Open source: yes
- AI role: core to the product
“Olmo-core 3 is Ai2's open training infrastructure for large mixture-of-experts language models”
What it works with
Names AstaBrief, DeepEP v2 and 3 more on its pages.
- Names: AstaBrief · DeepEP v2 · Megatron-Core · +2
“Open-sourcing AstaBrief, the fast report-generation model in Asta”
“We’ve also experimented with DeepEP v2, an alternative way of handling communication between experts across GPUs”
What it offers as proof
Live, no waitlist. Claims “47-billion-parameter MoE reached 52,000 tokens per second per GPU”. Instant signup.
- Number claimed: 47-billion-parameter MoE reached 52,000 tokens per second per GPU
- Instant signup: yes
“Log In Sign Up”
“In one benchmark, a 47-billion-parameter MoE reached 52,000 tokens per second per GPU”
Sources
Tablif read these pages, last on 3 Oct 2026. Everything above is what the product says about itself; Tablif does not verify every claim.
Similar products in Tablif Town
- Tensorant: Fine-tuning and LLM Inference Provider
- GOSH.AI DePools: Run the top open-source model at the best possible price
- Prime Inference: Fast, reliable serving for frontier open models
- SursumAI: Deploy models on your own machine
- Kolibri: Europe's newest powerful open-source AI model
- Limoni TUI Engine: Ultra-fast, zero-allocation TUI framework for Go
- GPT-Synopsys: OpenAI and Synopsys partner on AI for chip design
- OpenCreative: Open-source alternative for Higgsfield and ElevenLabs