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

What are some of the latest llm models released?

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.0-flash-grounding’s answer.
  1. User

    What are some of the latest llm models released?

  2. Perplexity Assistant

    Recent LLM releases showcase advancements in efficiency, multimodal capabilities, and open-source accessibility. Below are key models from late 2024 to early 2025:


    Proprietary Models

    Grok-3 (xAI)

    • Release: February 17, 2025
    • Parameters: Unknown (predecessor Grok-1 had 314B)
    • Features:
      • Real-time integration: Directly linked to X (Twitter) data for up-to-date responses.
      • Context window: 1 million tokens (vs. 128k in earlier versions).
      • Modes: "Think" (reasoning), "Big Brain," and "Deep Search" for specialized tasks[[3]][[6]].

    Gemini 2.0 Flash/Pro (Google)

    • Release: December 2024
    • Features:
      • Speed: Twice as fast as Gemini 1.5 with lower latency.
      • Cost: Over 50% price reduction for Pro versions[[1]][[6]].
      • Multimodal: Enhanced text-image processing via DataGemma, which integrates statistical data via Retrieval Interleaved Generation (RIG)[[1]].

    Claude 3.7 Sonnet (Anthropic)

    • Release: Early 2025
    • Features:
      • Extended thinking mode: Self-reflection loops for iterative reasoning.
      • Coding: Strong performance in front-end web development and problem-solving[[7]].

    Open-Source Models

    Llama 3.3 (Meta)

    • Release: December 2024
    • Parameters: 70B (multimodal)
    • Features:
      • Context window: 128k tokens.
      • Multimodal: Processes text and images for tasks like chart interpretation[[7]].

    Mistral Small 3

    • Release: January 2025
    • Parameters: 24B
    • Features:
      • Latency: Processes ~150 tokens/second (3x faster than Llama 3.3 70B).
      • License: Apache 2.0, deployable on low-resource devices[[7]].

    DeepSeek-R1

    • Release: January 2025
    • Parameters: 671B
    • Features:
      • Training: Reinforcement-learned from DeepSeek-V3, no pretraining.
      • License: MIT[[6]].

    Enterprise-Focused Models

    Inflection-3.0 (Inflection AI)

    • Release: October 2024
    • Features:
      • Efficiency: Achieves ~94% of GPT-4’s performance with 40% fewer training resources[[3]].

    Nova (Amazon)

    • Release: December 2024
    • Features:
      • Variants: Nova Micro, Lite, and Pro for diverse use cases[[6]].

    Notable Trends

    • Real-time data: Grok-3 and Gemini 2.0 prioritize live information integration.
    • Efficiency: Mistral Small 3 and Inflection-3.0 optimize for low-resource deployment.
    • Multimodal: Llama 3.3 and Gemini 2.0 expand text-image capabilities.

    For developers, open-source options like Llama 3.3 and Mistral Small 3 offer flexibility, while proprietary models like Grok-3 and Claude 3.7 target specialized enterprise needs.

  3. User

    Please sort the mode releases chronically

  4. Perplexity Assistant

    Here are notable LLM releases sorted chronologically based on available data:


    2024 Releases

    January

    • Qwen2.5 (Alibaba Cloud): Multimodal models (0.5B–72B) with 1M token context[[4]][[6]].
    • DeepSeek-R1 (DeepSeek): 671B reasoning model via reinforcement learning[[1]][[2]].

    February

    • Gemini 2.0 Flash/Pro (Google): Enhanced speed and multimodal capabilities[[1]][[6]].
    • Nova (Amazon): Enterprise-focused models (Micro, Lite, Pro)[[1]][[2]].

    March

    • Claude 3.5 Sonnet (Anthropic): Extended reasoning and coding capabilities[[3]][[6]].
    • Llama 3.3 (Meta): 70B multilingual text model[[3]][[8]].

    September

    • o1-preview/o1-mini (OpenAI): Reasoning-focused models for STEM tasks[[7]].

    December

    • DeepSeek-V3 (DeepSeek): 671B base for R1[[2]][[6]].
    • Gemini 2.5 Pro (Google): Experimental "thinking model" with 1M token context[[5]][[6]].

    2025 Releases

    January

    • DeepSeek-R1 (DeepSeek): Open-source reasoning model[[1]][[2]].
    • Sonus (Rubik's AI): Successor to Nova[[2]].

    February

    • GPT-4.5 (OpenAI): Largest unsupervised model (128K context)[[1]][[7]].
    • Claude 3.7 Sonnet (Anthropic): Hybrid reasoning with customizable "thinking" mode[[1]][[4]].
    • Grok-3 (xAI): 1M token context and real-time data integration[[3]][[5]].

    March

    • Gemini 2.5 Pro (Google): Official release of "thinking model"[[5]][[6]].

    Upcoming Models

    • GPT-5 (OpenAI): Anticipated mid-2025[[1]][[3]].
    • Claude 4 (Anthropic): Next-generation model[[3]].

    For open-source options, DeepSeek-R1 and Llama 3.3 remain prominent, while proprietary models like GPT-4.5 and Claude 3.7 target specialized enterprise needs.

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