Blog

Google DeepMind releases Gemini 2.0 Flash, a multimodal AI model designed for the agentic era with native image generation and double the speed of previous versions.

DeepSeek's latest upgrade to the V2.5 endpoint delivers significant jumps in mathematical reasoning and coding benchmarks, setting a new bar for MoE efficiency.

Meta releases Llama 3.3, a 70B parameter open-source model matching 405B performance with massive efficiency gains for developers.

OpenAI launches o1-Pro, a high-compute reasoning model available on ChatGPT Pro, designed to solve complex technical tasks with enhanced accuracy.

Amazon unveils the Nova family on AWS Bedrock. Optimized for enterprise, multimodal, and high-performance reasoning.

Alibaba Cloud launches Qwen2.5-Coder with a 32B variant matching GPT-4o coding ability, available under Apache 2.0 license.

Mistral AI releases Pixtral Large, a 124B parameter multimodal model with 128K context and open weights, challenging closed-source giants.

Tencent releases Hunyuan-Large, the largest open-source MoE model. 389B params, 256K context, beats Llama 3.1.

Anthropic launches Claude Haiku 3.5 with 200K context, $0.80 input pricing. Ideal for high-volume AI tasks.