Arcee AI Releases Trinity Large Preview: 400B-Parameter MoE Model with 512K Context Window
Arcee AI has released Trinity Large Preview, a 400B-parameter sparse Mixture-of-Experts model with 13B active parameters per token using 4-of-256 expert routing. The model supports context windows up to 512K tokens and is available with open weights under permissive licensing.
Trinity Large Preview — Quick Specs
Arcee AI Releases Trinity Large Preview: 400B-Parameter MoE Model with 512K Context Window
Arcee AI has released Trinity Large Preview, a 400B-parameter sparse Mixture-of-Experts (MoE) language model with 13B active parameters per token. The model uses 4-of-256 expert routing architecture and is currently available for free through OpenRouter.
Technical Specifications
Trinity Large Preview features:
- Total parameters: 400 billion (sparse)
- Active parameters per token: 13 billion
- Expert routing: 4-of-256 MoE architecture
- Context window: Up to 512K tokens (native support)
- Current deployment: 128K context window using 8-bit quantization
- Pricing: Free during preview period (ends April 22, 2026)
- License: Open weights with permissive licensing
Capabilities and Target Use Cases
According to Arcee AI, Trinity Large Preview excels in creative writing, storytelling, role-play, chat scenarios, and real-time voice assistance. The company claims the model performs better in these areas than typical reasoning models.
The model was specifically trained for agentic workflows, designed to navigate agent frameworks including OpenCode, Cline, and Kilo Code. Arcee AI states it handles complex toolchains and long, constraint-filled prompts effectively.
Deployment Details
The Preview API currently serves the model at 128K context using 8-bit quantization for practical deployment, though the architecture natively supports context windows up to 512K tokens. The model is available through OpenRouter's API with standard OpenAI-compatible formatting.
Benchmark scores have not been disclosed. The free preview period will end on April 22, 2026, though future pricing has not been announced.
What This Means
Trinity Large Preview represents Arcee AI's entry into frontier-scale models with a focus on efficiency through sparse MoE architecture. The 400B total parameter count with only 13B active per token aims to deliver large model capabilities at lower computational cost. The open weights and permissive licensing lower barriers for developers and researchers to experiment with frontier-scale models, particularly for agentic applications. However, the lack of published benchmark scores makes it difficult to assess performance against competing models in the 400B+ parameter class.
Related Articles
Inception launches Mercury 3 diffusion LLM: 1M context, 1,068 tokens/sec, $0.20/$0.70 per 1M tokens
Inception released Mercury 3, a diffusion-based LLM with a 1M-token context window that it says runs at 1,068 tokens per second on NVIDIA GPUs. Pricing is $0.20 per 1M input and $0.70 per 1M output tokens, with a 50% launch discount. Inception claims quality on par with or better than GPT-6 Luna and Claude Haiku 5.5 at 4–8x the speed.
MiniMax M3.1 Flash scores 42 on AA Intelligence Index at $0.15/$0.60 per 1M tokens, open weights Oct 23
MiniMax has released M3.1 Flash, a reasoning model that scores 42 on the Artificial Analysis Intelligence Index, up from 29 for M3. It costs $0.15 per 1M input tokens and $0.60 per 1M output tokens, and open weights are planned for October 23.
Mistral releases Timestral-209M, a time series model claiming top zero-shot scores on three benchmarks
Mistral AI has released Timestral-209M, the largest of four time series foundation models (5M, 28M, 123M, 209M parameters). According to the model card, it leads GIFT-Eval, TIME and fev-bench in zero-shot forecasting, with a context of up to 32,768 time steps and 21-quantile probabilistic output.
Tencent releases Hy-Memory-30B-A3B, a 3B-active model that writes agent memory for Mem0 and MemOS
Tencent Hunyuan released Hy-Memory-30B-A3B, a 30B-parameter mixture-of-experts model with about 3B active parameters, built to run the write path in external memory frameworks. Tencent reports a 55.49 cross-harness average, ahead of Claude Opus 5 (54.62) and DeepSeek V4.1 Flash (53.64) on its own fixed 20% benchmark subsamples.
Comments
Loading...