Ricursive releases Ripple-65B Preview, a general-purpose variant of its AI model for automating advanced chip design, under Apache 2.0, costing $0.60/1M output

This model achieves state-of-the-art performance on coding tasks at a fraction of the typical training cost, making advanced AI capabilities more accessible.

Ricursive Intelligence has released Ripple-65B Preview, an open-weight AI model designed for general-purpose tasks, including advanced chip design automation. This model, trained on 12 trillion tokens with a context length of up to 1 million, sets a new U.S. open-weight state-of-the-art on agentic coding benchmarks like TerminalBench 2.1, surpassing significantly larger models. The company developed and trained Ripple-65B Preview for $32 million, a fraction of the cost typically associated with frontier AI models. Its dense architecture is optimized for efficiency on NVIDIA hardware, enabling strong reasoning performance and low-latency generation, and can run on local devices. The model is available on multiple platforms including Hugging Face and Together AI, with pricing at $0.60 per 1 million output tokens.

Ricursive's achievement of U.S. open-weight SOTA on agentic coding with Ripple-65B Preview, particularly its claimed $32 million training cost, signals a potential paradigm shift in LLM development. This contrasts sharply with the multi-billion dollar figures often cited for training frontier models, suggesting that significant performance gains may be achievable through architectural innovations and efficient training methodologies rather than sheer scale and brute-force compute. The model's efficiency on local hardware also hints at a future where powerful AI can be deployed more widely without reliance on massive data centers.

The release of Ripple-65B Preview under Apache 2.0 license democratizes access to a highly capable agentic coding model, impacting researchers and developers who can now leverage SOTA performance without the prohibitive costs of training. This move could accelerate innovation in areas requiring complex code generation and chip design automation. The focus on efficient training and serving also sets a new benchmark for cost-effectiveness in the competitive AI landscape, potentially pressuring other providers to re-evaluate their own development and pricing strategies.