447 results. Find your next favorite, with a little help from the community.
OpenAI's GPT-3.5 model series (text-davinci-003, gpt-3.5-turbo) introduced in late 2022; the fine-tuned lineage that powered the initial ChatGPT launch.
Meta's 70B dense Instruct-only open-weight LLM from Llama 3.3 with a 128K token context, released December 2024 under the Llama 3.3 Community License.
754B MoE (glm_moe_dsa) with DSA sparse attention, MIT license; iterative successor to GLM-5 for long-horizon agentic tasks with thousands of tool calls.
OpenAI's smaller open-weight 20B-parameter reasoning model released under Apache 2.0, designed for on-device/self-hosted use with function calling and structured outputs.
Mistral's multimodal OCR API for document understanding; parses text, tables, equations, and images from PDFs and scans into structured markdown output.
Anthropic's Claude Opus 4, initial Claude 4 flagship for coding, reasoning, and agent workflows; retired June 2026 in favor of Opus 4.8.
First Generative Pre-trained Transformer from OpenAI (2018), a 117M-parameter decoder-only model pre-trained on unlabeled text then fine-tuned for NLP tasks.
OpenAI's second-generation reasoning model with tool use, delivering gains on math, coding, and science benchmarks over o1; released alongside o4-mini in April 2025.
685B/37B-active MoE with 256 experts, 128K context, MIT license; debuts DeepSeek Sparse Attention (DSA) for cheap long-context training and inference.
Enhanced parallel-thinking mode for Gemini 2.5 Pro announced at I/O 2025; an advanced version later reached gold-medal standard at IMO 2025.
Smallest Gemini model, designed for on-device inference in Android (starting on Pixel 8 Pro) and Chrome, enabling local AI features without cloud calls.
No public release from xAI; the current post-Grok 4 line is Grok 4.1 (Nov 2025) and Grok 4.1 Fast (Nov 2025), with no Grok 4.3 or Grok 4.20 announced.
Hybrid MoE (671B/37B active) with toggleable thinking mode, 128K context, MIT license; unifies V3 chat and R1 reasoning in a single hybrid model.
Refreshed 3.8B Phi-3.5-mini-instruct SLM from Microsoft with a 128K context window and multilingual gains, released on Hugging Face under an MIT license.
Open-source MCP server that lets any LLM control Blender 3D via a Python addon, exposing scene manipulation, modeling, and rendering commands.
MCP server from Bright Data giving AI agents web search, scraping, and browser automation via Bright Data's unlocker/proxy infrastructure.
Anthropic's Claude Opus 4.1, an upgrade to Opus 4 for agentic tasks and coding; scored 74.5% on SWE-bench Verified. Same pricing as Opus 4.
AI-facing directory of Claude Skills and MCP servers where agents can browse, test, and rate entries; awards a blue badge to those that run successfully.
3B vision-language OCR model (deepseek_vl_v2) with dynamic-resolution tiling and layout grounding, Apache-2.0; second-gen Visual Causal Flow architecture.
236B/21B-active MoE with Multi-head Latent Attention, 128K context, DeepSeek license; pretrained on 8.1T tokens with 93% KV cache reduction over dense peers.
Parallel-reasoning mode built on Gemini 3 Pro for hard math, science, and logic problems; available to Google AI Ultra subscribers in the Gemini app.
32B dense bilingual GLM-4 model with 32K context (YaRN extendable), MIT license; strong Chinese/English chat and function-calling foundation model.
OpenAI's GPT-5.1 update with Instant and Thinking variants, delivering a warmer conversational tone and better instruction following on top of GPT-5's architecture.
OpenAI's image model succeeding gpt-image-1 with more precise edits, better instruction following, and improved rendering of dense small text; in API and ChatGPT.