Skip to content
TheChatArchiveThe Chat Archive
Perplexitysonar-reasoning· PerplexityHugging Face LMArena Search Arena 24K

what are the new and relevant Large Language Models released in April/2025?

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.5-pro-exp-03-25-grounding’s answer.
  1. User

    what are the new and relevant Large Language Models released in April/2025?

  2. Perplexity Assistant

    As of April 2025, the search results do not explicitly mention any new Large Language Models (LLMs) released specifically in April. However, recent advancements and notable models from early 2025 include:


    Recent LLM Developments (Early 2025)

    1. GPT-4.5 "Orion" (OpenAI)

    Released in February 2025, this model focuses on efficiency and multimodal capabilities:

    • Reduced energy consumption (targeting 30% reduction via hardware optimization and quantization)[[1]].
    • Advanced multimodal integration (joint embedding spaces for text, images, and audio)[[1]].
    • Enhanced reasoning through improved training data and symbolic AI integration[[1]].
    • Context window: 128,000 tokens[[6]].

    2. DeepSeek R1

    Launched in January 2025, this open-source model excels in math and coding:

    • Parameters: 671B total (37B active)[[6]].
    • Performance: Matches or outperforms OpenAI’s reasoning models (e.g., MATH-500, AIME 2024)[[6]].
    • Cost efficiency: Low training costs compared to competitors[[6]].

    3. Claude 3.7 Sonnet (Anthropic)

    Introduced in early 2025, this hybrid model combines deep reasoning with transparent workflows:

    • Scratchpad feature: Visualizes reasoning steps for user refinement[[5]].
    • Interpretable logic: Adjustable reasoning levels for tasks like research or content creation[[5]].

    Emerging Frameworks and Trends

    While no April-specific models are cited, notable advancements include:

    • Chain-of-Agents (Google): A framework enabling LLM collaboration for long-context tasks (e.g., summarization, code completion), introduced in January 2025[[4]].
    • Multimodal and specialized models: Continued focus on integrating video, audio, and domain-specific fine-tuning (e.g., GPT-4V for multimodal tasks)[[3]][[6]].

    Key Players in 2025

    ModelDeveloperRelease DateNotable Features
    GPT-4.5OpenAIFeb 2025Energy efficiency, multimodal integration
    DeepSeek R1DeepSeekJan 2025Math/coding focus, open-source
    Claude 3.7 SonnetAnthropicEarly 2025Hybrid reasoning, scratchpad visualization
    PaLM 2GoogleMay 2023340B parameters, powers Gemini

    For April 2025, the emphasis appears to be on refining existing models (e.g., GPT-4.5) and advancing frameworks like Chain-of-Agents rather than releasing entirely new architectures[[1]][[4]][[6]].

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.