Updated continuously · 2026

Changelog

Every AI model added, benchmark refreshed, and tool shipped. We track frontier releases (Claude, GPT, Gemini, DeepSeek, Qwen, Mistral) and open-weight launches as they happen — typically within 7 days of the official release.

📅 Published: May 9, 2026🔄 Last Updated: May 9, 2026✓ Manually Reviewed

Mistral Medium 3.5 added

New Model

Mistral AI shipped Mistral Medium 3.5 on April 30 — the first unified Magistral (reasoning) + Pixtral (vision) + Devstral (code) model. Full benchmark coverage and self-hosting guide live.

DeepSeek V4-Pro and V4-Flash benchmarks added

Benchmark Update

DeepSeek released V4 on April 24, 2026. V4-Pro (1.6T MoE / 49B active) hits 76.4% SWE-Bench Verified — top open-weight result. MIT licensed. Full deployment guide for vLLM, SGLang, and Mac Studio M3 Ultra.

Qwen3-Coder-Next and Qwen3.6-27B added

New Model

Alibaba shipped Qwen3-Coder-Next (80B / 3B active, 70.6% SWE-Bench) and Qwen3.6-27B (dense 27B beating 397B MoE). Both Apache 2.0. Now the highest-scoring open coding model.

Kimi K2.6 added

New Model

Moonshot AI shipped Kimi K2.6 (1T MoE / 32B active) on April 20. Modified MIT license. Strong reasoning and Chinese-language capability; 200K context.

Claude Opus 4.7 and Claude Sonnet 5 added

New Model

Anthropic shipped Claude Sonnet 5 (codename "Fennec", 92.4% SWE-Bench Verified — current SOTA on agentic coding) and Claude Opus 4.7 (Adaptive Thinking, 95.2% AIME 2025). Both available via API.

GPT-5.5 family added (Pro / Standard / Mini)

New Model

OpenAI shipped GPT-5.5 in three tiers. GPT-5.5 Pro hits 96.7% AIME 2025 — top math/competition score. Full pricing breakdown, API setup, and head-to-head benchmarks vs Claude.

GLM-5 added — Zhipu 745B/44B active MoE

New Model

Zhipu AI shipped GLM-5 (745B total / 44B active, MIT license). Runs on Huawei Ascend hardware. Strong reasoning at lower deployment cost than DeepSeek V3.

Gemini 3.1 Pro added — Google DeepMind

New Model

Google DeepMind shipped Gemini 3.1 Pro on February 5, 2026. 1M context window, 77.1% ARC-AGI-2 (highest published), Deep Think mode. Full benchmark coverage and API guide live.

🎯
AI Learning Path

Go from reading about AI to building with AI

10 structured courses. Hands-on projects. Runs on your machine. Start free.

Model creators: spot a mistake?

If you work at one of the labs covered above and a benchmark or capability claim is wrong on our model pages, email us. We update within 7 days — typically same week if the change is documented in your model card or technical report. Independent coverage that respects the facts, not marketing fluff.

For a higher-level overview of how the 2026 AI landscape compares head-to-head, see the AI Model Leaderboard and the Best AI Models 2026 pillar.

Subscribe to updates

PR

Written by Pattanaik Ramswarup

Creator of Local AI Master

I build Local AI Master around practical, testable local AI workflows: model selection, hardware planning, RAG systems, agents, and MLOps. The goal is to turn scattered tutorials into a structured learning path you can follow on your own hardware.

✓ Local AI Curriculum✓ Hands-On Projects✓ Open Source Contributor
Free Tools & Calculators