Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-pro-exp-03-25-grounding· GoogleHugging 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 Perplexitysonar-reasoning’s answer.
  1. User

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

  2. Gemini Assistant

    As of early April 2025, the field of Large Language Models (LLMs) continues to evolve rapidly[1][2][3]. Based on recent announcements and reports, here are some new and relevant LLMs mentioned:

    1. Llama 4 (Meta)

    • Meta announced the release of its latest generation of LLMs, the Llama 4 series, in early April 2025[4][5].
    • Llama 4 Scout and Llama 4 Maverick: These are described as Meta's first open-weight, natively multimodal models (handling text, images, video, audio). They feature Mixture-of-Experts (MoE) architecture for efficiency[4].
      • Scout: Has 17 billion active parameters (16 experts) and a large 10 million token context window[5]. It's designed to be powerful yet compact enough to run on a single high-end GPU (Nvidia H100)[4]. Meta claims it outperforms models like Google's Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 on various benchmarks[4].
      • Maverick: Larger model (17 billion active parameters, 128 experts, 400 billion total parameters) positioned to compete with OpenAI's GPT-4o and Google's Gemini 2.0 Flash. Meta states it achieves results comparable to DeepSeek-V3.1 on coding/reasoning tasks using fewer active parameters.
    • Llama 4 Behemoth: Previewed as a highly powerful model (2 trillion parameters, 288 billion active), described by Mark Zuckerberg as potentially "the most powerful base model in the world."[4] It is still in development and uses MoE architecture, expected to excel in STEM benchmarks[4].
    • Availability: Scout and Maverick are available for download via Llama.com and Hugging Face[4][5]. More details are expected at Meta's LlamaCon event on April 29, 2025[4][5].

    2. Other Relevant Models (Mentioned in April 2025 Context)

    While not all released precisely in April, several other models are frequently discussed as the leading or most relevant LLMs currently:

    • GPT-4o (OpenAI): Released mid-2024, this remains a flagship model known for its "omni" capabilities across multiple modalities[2].
    • Claude 3.7 Sonnet (Anthropic): Released February 2025, noted for its hybrid reasoning capabilities (switching between fast and extended thinking modes) and strong performance in logic, math, and coding[2].
    • Gemini 2.0 Flash (Google DeepMind): Unveiled early 2025 as part of the Gemini 2.0 family, this is Google's flagship "agentic" LLM focused on performing actions and native tool use, featuring a large context window (potentially 1 million tokens for the broader Gemini 2.5 family). Google's Gemma 3 (open model) is also relevant.
    • Grok 3 (xAI): Introduced early 2025 by Elon Musk's xAI, positioned as a competitor to top models like GPT and Claude, emphasizing scale and rapid iteration[2].
    • DeepSeek R-1 (DeepSeek): An open-source model released by the Chinese startup DeepSeek in 2025, gaining attention for achieving high reasoning performance (comparable to proprietary models) with potentially fewer resources[2][6].
    • Qwen Models (Alibaba): Alibaba released Qwen2.5-Max in early 2025 and QwQ-32B in March 2025[3]. The QwQ model is highlighted for mathematical reasoning and coding efficiency[2][3].

    3. OpenAI's Upcoming Open-Source Model[7][8]

    • OpenAI announced in early April 2025 its plan to release a new open-source (or "open-weight") LLM "in months"[8]. This would be their first open release since GPT-2 in 2019[7][8]. It's expected to feature powerful reasoning capabilities and be runnable on local hardware[7].

    4. LLM Tooling Updates

    • LLM 0.24: A command-line tool and Python library for interacting with LLMs received an update (version 0[9].24) in early April 2025, adding features like "fragments" to better support the increasingly long context windows of modern models like Llama 4 Scout[9].
    • MLPerf Inference v5.0: Released in April 2025, this benchmark suite from MLCommons added new tests specifically for large LLMs, including benchmarks using Llama 3.1 405B Instruct (a very large model) and Llama 2 Chat 70B (for low-latency interactive tasks), reflecting the need to measure performance on current, demanding models[10].

    In summary, April 2025 has seen a significant release from Meta with its Llama 4 series, pushing boundaries in multimodality and context length for open models[4][5]. Alongside this, discussions continue around recently released powerful models from OpenAI, Google, Anthropic, xAI, and DeepSeek, highlighting the intense competition and innovation in the LLM space. OpenAI's planned return to open-source releases is also a notable development this month.

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