News & Changelog
Latest AI model releases, feature updates, and industry news — all in one feed.
New models: Qwen3.8-Flash-Next
Added Alibaba's Qwen3.8-Flash-Next (Aug 26, 2026), an open-weight 125B mixture-of-experts model with 6B active parameters per token that opens a new Qwen Flash line on the tracker.
New models: GPT-6 Astra, Claude Opus 5, Gemini 3.8 Flash, Grok 4.6 & more
Catch-up update covering 10 June to 8 September 2026: Anthropic's Claude Sonnet 5 (Jun 30), Opus 5 (Jul 24) and Fable 5.1 (Sep 1); OpenAI's GPT-5.6 Sol, Terra and Luna (Jul 9) and GPT-6 Astra (Sep 3); Google's DiffusionGemma (Jun 10), Gemini 3.6 Flash and 3.5 Flash-Lite (Jul 21), 3.7 Flash (Aug 13) and 3.8 Flash (Sep 2); xAI's Grok 4.5 (Jul 8) and 4.6 (Aug 12); Meta's Muse Spark 1.1 (Jul 9), 1.2 (Aug 5), Muse Glimmer 30B (Aug 10) and Muse Spark 1.3 (Sep 2); Mistral's Leanstral 1.5 (Jul 2); Alibaba's Qwen 3.8-Max (Aug 3), Qwen3.8-2.4T-A95B (Aug 12) and Qwen3.8-27B (Aug 14); and DeepSeek's V4-Flash-0731 (Jul 31) and V4-Pro-0813 (Aug 13).
New models: Claude Fable 5, MAI-Code-1-Flash, Qwen 3.7-Plus & more
Added Anthropic's Claude Fable 5 (Jun 9) — the first generally available Mythos-class model — plus Microsoft's MAI-Code-1-Flash (Jun 2), Qwen 3.7-Plus (Jun 1), Gemma 4 12B Unified (Jun 3), Qwen3.6-27B, Qwen3-Coder-Next and Mistral's Leanstral. The proprietary Qwen Max & Plus line was also reorganized.
Relaunched as Next AI Model
The tracker is now live at www.nextaimodel.com with a focused homepage: release forecasts, what just dropped, and a weekly newsletter. The old llmdrops.com domain redirects here.
Data-driven release cards
Each recent release now shows a compact comparison with the version it replaces: benchmark scores before → after, price and context-window changes — measurable facts only, with sources linked.
Honest overdue forecasts
When a model passes its typical release cycle we now mark it Overdue and show how many days late it is, instead of displaying a predicted date stuck in the past.
Release archive
The new Archive view lists every release we track, newest first, with plain-language summaries that get backfilled over time — the full history behind the forecasts.
New model family: Qwen 3.5 Medium
Added Alibaba Qwen 3.5 Medium (Feb 24, 2026) as a new family. The series includes four open-source models (Qwen3.5-27B, 35B-A3B, 122B-A10B, Flash) offering near-flagship performance on local hardware under Apache 2.0.
#1 SVG Animated Car · 5 models
Gemini 3.1 Pro (AI Studio), Gemini 3.1 Pro (Gemini App), Claude Opus 4.6, GPT 5.2 Thinking (xhigh), Grok 4.20
We gave the same prompt to 5 leading AI models: "Generate an SVG animation of a red car on a highway, generating smog behind it, and clouds slowly moving in the sky." Click on each result to see the full animation. Judge for yourself!
News & Benchmark tabs added
The site now has a dedicated News section (you're looking at it!) with a changelog, AI world news, and a Benchmark tab with a 'Coming Soon' teaser.
New models: Gemini 3.1 Pro & Grok 4.20
Added Google Gemini 3.1 Pro (Feb 19) and xAI Grok 4.20 (Feb 17) to the tracker. Both are now showing as 'Just Updated'.
Google announces Gemini 3.1 Pro with native multimodal reasoning
Google released Gemini 3.1 Pro, their most capable model yet, featuring improved long-context understanding up to 2M tokens and real-time audio/video processing. Available through Google AI Studio and Vertex AI.
MacRumors-style cycle badges
Inspired by MacRumors Buyer's Guide, each model now shows a cycle phase badge: Just Updated, Early Cycle, Mid-Cycle, Expected Soon, or Overdue. Colors updated for better readability.
New models: Claude Sonnet 4.6 & Qwen 3.5
Added Anthropic Claude Sonnet 4.6 (Feb 17) and Alibaba Qwen 3.5 (Feb 16) to the tracker.
xAI releases Grok 4.20 with improved coding abilities
Elon Musk's xAI launched Grok 4.20, claiming significant improvements in code generation and mathematical reasoning. The model is available to X Premium+ subscribers and through the xAI API.
EU AI Act enforcement begins for general-purpose AI
The European Union started enforcing the AI Act provisions for general-purpose AI models. Companies now need to provide technical documentation, comply with transparency requirements, and undergo evaluations for systemic risk models.
Show all predictions & Analytics
Added expand/collapse button to see all 13 predictions in the Coming Soon section. Integrated Vercel Analytics for visit tracking.
Anthropic releases Claude Sonnet 4.6 with extended thinking
Anthropic launched Claude Sonnet 4.6, featuring improved extended thinking capabilities and better instruction following. The model powers Claude Code and is available through the API.
Site rebranded to LLM Drops
Rebranded to 'LLM Drops' with the new domain llmdrops.com. Updated all meta tags and social cards.
OpenAI reportedly training GPT-5.5 on custom hardware
According to multiple reports, OpenAI has begun training a new model generation (rumored GPT-5.5) using a mix of custom AI accelerators and NVIDIA H200 clusters. Release is expected mid-2026.
Added Qwen & DeepSeek providers
Expanded coverage to 8 providers with the addition of Alibaba (Qwen) and DeepSeek model families, including their coder variants.