Beam
Beam is Reflection’s first open-weight model, a sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active parameters, built for coding, reasoning, and agentic workloads. Its capabilities come from large-scale pretraining and reinforcement learning, including training on 23.8 trillion diverse, curated, high-quality tokens from the web, public sources, and proprietary licensed datasets. Beam was trained with a particular focus on coding and agentic performance and is designed to deliver competitive open-weight capabilities with efficient inference compute. It supports complex software engineering, terminal, STEM, web search, tool-use, and general knowledge tasks, with reinforcement learning designed to improve multi-step reasoning, tool use, and adaptation to environment feedback. Users can control the tradeoff between performance and token usage through a reasoning effort parameter.
Learn more
Phi-4-reasoning
Phi-4-reasoning is a 14-billion parameter transformer-based language model optimized for complex reasoning tasks, including math, coding, algorithmic problem solving, and planning. Trained via supervised fine-tuning of Phi-4 on carefully curated "teachable" prompts and reasoning demonstrations generated using o3-mini, it generates detailed reasoning chains that effectively leverage inference-time compute. Phi-4-reasoning incorporates outcome-based reinforcement learning to produce longer reasoning traces. It outperforms significantly larger open-weight models such as DeepSeek-R1-Distill-Llama-70B and approaches the performance levels of the full DeepSeek-R1 model across a wide range of reasoning tasks. Phi-4-reasoning is designed for environments with constrained computing or latency. Fine-tuned with synthetic data generated by DeepSeek-R1, it provides high-quality, step-by-step problem solving.
Learn more
Sakana Fugu Ultra
Sakana Fugu Ultra is the higher-performance version of Sakana Fugu, built to coordinate a deeper pool of expert AI agents for demanding, high-stakes tasks. The model operates through a single OpenAI-compatible API while dynamically orchestrating multiple powerful models behind the scenes. It is designed to maximize answer quality for complex workflows such as coding, code review, paper reproduction, cybersecurity analysis, scientific reasoning, patent investigation, and autonomous research. Fugu Ultra uses learned orchestration techniques to assemble, route, and coordinate agents instead of relying on hand-designed workflows or a single frontier model. Users can access advanced multi-agent intelligence without manually managing separate models, prompts, or collaboration patterns. Sakana Fugu Ultra is built for teams that need stronger performance, deeper reasoning, and more reliable results on difficult multi-step problems.
Learn more
Bitext
Bitext provides multilingual, hybrid synthetic training datasets specifically designed for intent detection and LLM fine‑tuning. These datasets blend large-scale synthetic text generation with expert curation and linguistic annotation, covering lexical, syntactic, semantic, register, and stylistic variation, to enhance conversational models’ understanding, accuracy, and domain adaptation. For example, their open source customer‑support dataset features ~27,000 question–answer pairs (≈3.57 million tokens), 27 intents across 10 categories, 30 entity types, and 12 language‑generation tags, all anonymized to comply with privacy, bias, and anti‑hallucination standards. Bitext also offers vertical-specific datasets (e.g., travel, banking) and supports over 20 industries in multiple languages with more than 95% accuracy. Their hybrid approach ensures scalable, multilingual training data, privacy-compliant, bias-mitigated, and ready for seamless LLM improvement and deployment.
Learn more
