T5Gemma 2

« A fast and cost-effective model for code and long-running agent tasks, with clear improvements in multi-step reasoning. Available at the same price as the previous version through the end of 2026 »

« Moonshot's LLM model, which rivals the best proprietary models on numerous benchmarks. Designed for long-form coding, advanced reasoning, and knowledge work tasks, it is also the first open-source model to surpass the 2.8 trillion-parameter mark »

« A fast and cost-effective model designed for code, agents, and complex workflows, with a 1-million-token context and significant improvements on long-running tasks and multi-step execution. Compared to its predecessor, this model scores 9 points higher on FrontierCode 1.1 and 16 points higher on DeepSWE v1.1 »

« A powerful LLM model designed for coding, research, and long-running agentic tasks, featuring 500,000 tokens of context, tool calls, and fine-tuned reasoning effort. Available directly in Cursor and Grok Build »

« A premium model designed for everyday use, offering improved capabilities in coding, reasoning, and agent-based tasks, while remaining more affordable than the next tier up. Available on the Pro and Max plans at Anthropic »

« A multimodal model with 552B parameters, of which only 8B are active at the input stage, featuring native visual understanding and accessible via API for your applications. This open-source model uses a Causal Encoder-Decoder architecture to respond to your text and image queries »

« A long-term coding and agent model based on GLM-5.2, featuring enhanced training, a 1-million-token context, and always-active reasoning that can be adjusted across three levels. This LLM stands out for its performance in logical reasoning and semantic understanding. »

« Anthropic's latest high-end model, optimized for agentic code, long-running tasks, and critical business scenarios. It features self-checking of responses and a context window of one million tokens. It is also capable of automatically adjusting the depth of its reasoning based on the difficulty of the problem »