Updated weekly · v2026.07

LLM.FINDER

The fastest way to find the right AI model for your task, hardware, and budget.

This site covers text & language models— GPT, Claude, Llama, DeepSeek, Gemini, Qwen and more. Image, video & audio generation are a different category and not included.

24+
Models Tracked
12+
Benchmarks
14
Task Categories
96+
Model Weaknesses
{// what you can do}

Everything You Need to Pick Your Model

// latest news

Fresh From the Labs

We track every major release. Here's what's new in 2026.

JUN 2026

Ornith-1.0 Family Released

DeepReinforce open-sources coding models — 397B MoE beats Claude Opus 4.7 on Terminal-Bench 2.1. MIT license.

JUN 2026

Qwen3.7-Max: Alibaba Closes Its Flagship

First closed-weight Qwen ever. 1M context, reasoning-native agent model, Arena Elo 1,475. Qwen3.7-Plus adds vision.

MAY 2026

Claude Opus 4.8 — 4× Fewer False Completions

SWE-bench leader. Now with sub-agent orchestration, adaptive thinking, and 2.5× fast mode at 1/3 the cost.

APR 2026

MiMo-V2.5 Pro — Xiaomi's 1.02T Beast

Built a full SysY compiler in Rust via 672 tool calls. MIT license, 1M context, native multimodal.

APR 2026

DeepSeek V4 Pro — 1.6T MoE, 49B Active

Best open-source math and code model. MIT license. FP8 optimized for domestic hardware.

FEB 2026

GLM-5.2 — Zhipu Trains Entirely on Ascend

744B MoE with 40B active. Async RL for agentic coding. First major model trained without NVIDIA.

// bottom line

Stop Googling "Which AI Should I Use?"

We've done the research so you don't have to. 24+ models. 12+ benchmarks. Real weaknesses verified against whitepapers. And a smart recommender that finds your best model in seconds.

Enter the Dashboard