Release v3.33: [2026-06-21] - Version 3.33 - MiniMax Hailuo 02 video model, MiniMax Speech 2.6, Mistral Voxtral TTS 4B, Apple Core AI/Foundation Models at WWDC26, Aion 1.0 on-device SLMs, Gemini CLI shutdown/Antigravity CLI replacement, Amazon Bedrock AgentCore Harness GA, Databricks Genie One GA, Vercel eve agent framework, Omnigent meta-harness, GitKraken Kepler ADE, Zensar AgentMesh, Konecta Kolibri, Sedai AI Agent Optimization, GMI Cloud AgentBox, OpenRouter Fusion, GLM 5.2 on Fireworks, Gemma 4 on Cerebras, MCP 2026-07-28 stateless spec RC, new VS Code extensions, browser automation tools, AI safety updates

Release v3.33

Release v3.33: [2026-06-21] - Version 3.33 - MiniMax Hailuo 02 video model, MiniMax Speech 2.6, Mistral Voxtral TTS 4B, Apple Core AI/Foundation Models at WWDC26, Aion 1.0 on-device SLMs, Gemini CLI shutdown/Antigravity CLI replacement, Amazon Bedrock AgentCore Harness GA, Databricks Genie One GA, Vercel eve agent framework, Omnigent meta-harness, GitKraken Kepler ADE, Zensar AgentMesh, Konecta Kolibri, Sedai AI Agent Optimization, GMI Cloud AgentBox, OpenRouter Fusion, GLM 5.2 on Fireworks, Gemma 4 on Cerebras, MCP 2026-07-28 stateless spec RC, new VS Code extensions, browser automation tools, AI safety updates

Latest Changelog Entry

[2026-06-21] - Version 3.33 - MiniMax Hailuo 02 video model, MiniMax Speech 2.6, Mistral Voxtral TTS 4B, Apple Core AI/Foundation Models at WWDC26, Aion 1.0 on-device SLMs, Gemini CLI shutdown/Antigravity CLI replacement, Amazon Bedrock AgentCore Harness GA, Databricks Genie One GA, Vercel eve agent framework, Omnigent meta-harness, GitKraken Kepler ADE, Zensar AgentMesh, Konecta Kolibri, Sedai AI Agent Optimization, GMI Cloud AgentBox, OpenRouter Fusion, GLM 5.2 on Fireworks, Gemma 4 on Cerebras, MCP 2026-07-28 stateless spec RC, new VS Code extensions, browser automation tools, AI safety updates

Added

• MiniMax Hailuo 02 (Video Generation): 3x params vs predecessor, 4x training data, native 1080p, #2 Artificial Analysis Video Arena; released 2026-06-20.
  • Source: https://www.minimax.io/news/minimax-hailuo-02
• MiniMax Video-01 (Video Generation): First AI-native video model from MiniMax; 720p/25fps; text-to-video and image-to-video; up to 6s clips; released 2026-06-20.
  • Source: https://www.minimax.io/news/video-01
• MiniMax Speech 2.6 (TTS Proprietary): <250ms latency; Fluent LoRA; non-standard text handling; released 2026-06-20.
  • Source: https://www.minimax.io/news/minimax-speech-26
• Mistral Voxtral TTS (TTS Proprietary): 4B parameter TTS; 9 languages; 70ms latency; voice cloning from 3s audio; $0.016/1K chars; released 2026-06-18.
  • Source: https://mistral.ai/news/voxtral-tts
• Apple Core AI / Foundation Models (SLMs / On-Device): WWDC26 announcement; successor to Core ML; on-device inference + Private Cloud Compute; hybrid platform spanning on-device, PCC, and third-party models (Claude/Gemini); open-source Swift framework; Aion 1.0 Instruct (open weights, Edge Insider preview) and Aion 1.0 Plan (14B reasoning, 32K context, agentic, shipping in-box "in the coming months"); released 2026-06-20.
  • Source: https://developer.apple.com/documentation/coreai, https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/
• Aion 1.0 (SLMs): Microsoft on-device AI family; Aion 1.0 Instruct (small open-weights SLM for everyday tasks, Edge Insider preview, HuggingFace open-source July 2026) and Aion 1.0 Plan (14B reasoning model, 32K context, agentic workflows, shipping in-box Windows "in the coming months"); released 2026-06-02.
  • Source: https://blogs.windows.com/windowsdeveloper/2026/06/02/build-2026-furthering-windows-as-the-trusted-platform-for-development/
• Antigravity CLI (agy) (CLI Tools): Google's replacement for Gemini CLI; Go-based; async multi-agent workflows; free tier ~20 req/day; released 2026-06-18.
  • Source: https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/
• Vercel eve (Multi-Agent Platforms): Open-source agent framework; filesystem-first TypeScript; durable execution; sandboxed compute; human-in-the-loop; public preview 2026-06-17.
  • Source: https://vercel.com/blog/introducing-eve
• Omnigent (Multi-Agent Platforms): Databricks open-source meta-harness; composes multiple agents (Claude Code, Codex, Pi); stateful policies; session sharing; Apache 2.0; alpha 2026-06-13.
  • Source: https://www.databricks.com/blog/introducing-omnigent-meta-harness-combine-control-and-share-your-agents
• GitKraken Kepler (Agent Platforms): Agentic Development Environment; multi-agent orchestration; conflict resolver; branch intelligence; released 2026-06-15.
  • Source: https://www.gitkraken.com/blog/introducing-kepler-the-delivery-engine-for-agent-driven-development
• Zensar ZenseAI.AgentMesh (Agent Platforms): Enterprise agentic AI platform; 80+ pre-built agents; governed orchestration; EU AI Act alignment; GA 2026-06-19.
  • Source: https://www.prnewswire.com/news-releases/zensar-technologies-launches-zenseaiagentmesh-to-accelerate-enterprise-ai-adoption-at-scale-302792208.html
• Konecta Kolibri (Agent Platforms): Agentic AI orchestration platform; 80% pre-built use cases; FinOps dashboards; GA 2026-06-16.
  • Source: https://konecta.com/news-insights/konecta-launches-kolibri-an-agentic-platform-to-speed-up-enterprise-deployment-of-agentic-ai-and-end-pilot-purgatory
• Sedai AI Agent Optimization (Agent Platforms): Autonomous platform for AI agent optimization; governance, observability, smart routing; early access 2026-06-09.
  • Source: https://www.prnewswire.com/news-releases/sedai-launches-the-first-autonomous-platform-for-ai-agent-optimization-302792208.html
• GMI Cloud AgentBox (Agent Platforms): Marketplace for AI agents; 100+ models via single key; usage-based pricing; GA 2026-06-08.
  • Source: https://www.gmicloud.ai/en/blog/agentbox-is-live-the-whole-stack-for-production-ai-agents-in-one-place
• OpenRouter Fusion (API Providers): Model synthesis tool; combines multiple model outputs via judge; Fable 5 + GPT-5.5 fusion scored 69.0% surpassing individual models; launched 2026-06-12.
  • Source: https://openrouter.ai/blog/announcements/fusion-beats-frontier/
• Kilo for GitHub (CLI Tools): AI coding agent living in GitHub; @kilocode-bot mentions in issues/PRs; Cloud Agent backend; released 2026-06-18.
  • Source: https://blog.kilo.ai/p/introducing-kilo-for-github
• Qoder 1.0 (IDEs): Alibaba Cloud Autonomous Development Desktop; Quest window for agent command; editor + agent parallel workspaces; GA 2026-06-16.
  • Source: https://www.alibabacloud.com/blog/introducing-qoder-1-0-from-ai-ide-to-autonomous-development-desktop_603260
• OpenHands Agent Canvas (Agent Platforms): Workspace for creating automations integrating Slack, GitHub; self-hostable; released 2026-06-16.
  • Source: https://www.openhands.dev/blog/introducing-agent-canvas
• Hermes Agent v0.17.0 (CLI Tools): Major desktop app update; live subagent watch-windows; VS Code theme support; released 2026-06-19.
  • Source: https://github.com/NousResearch/hermes-agent/releases/tag/v2026.6.19
• Mastra Harness (Multi-Agent Platforms): Agent loop abstraction; persistent sessions; tool approval; model switching; available in @mastra/core@1.43.0; released 2026-06-18.
  • Source: https://mastra.ai/blog/announcing-agent-harness
• Kling 3.0 Turbo (Video Generation): Fast-preview mode; 1-15s clips at 480p/720p; rapid iteration before full Kling 3.0 production; released 2026-06-17.
  • Source: https://news.financenewsworld.com/story/564950/kling-30-turbo-released-kling-ai-brings-fast-video-previews-for-rapid-creative-iteration.html
• ZONOS2 (TTS Open Source): Zyphra 8B MoE TTS; Apache 2.0; multilingual + code-switched; zero-shot voice cloning; ECAPA-TDNN speaker embeddings; released 2026-06-12.
  • Source: https://www.zyphra.com/our-work/zonos2
• dots.tts (TTS Open Source): RedNote 2B fully continuous AR TTS; 48kHz AudioVAE; no discrete tokens; Apache 2.0; released 2026-06.
  • Source: https://github.com/rednote-hilab/dots.tts
• Soniox v5 Async (STT): Structured speech-to-text; speaker separation; language IDs; normalized structured entities; released 2026-06-11.
  • Source: https://soniox.com/blog/soniox-v5-async
• Gnani Prisma v2.5 (STT): #1 in 8 of 9 Indian languages; 15% lower WER rural Hindi; 14M hours training data; released 2026-06-19.
  • Source: https://cxotoday.com/media-coverage/gnani-ai-launches-prisma-v2-5-ranked-1-in-8-of-9-indian-languages-on-real-world-and-noisy-asr-benchmarks/
• NVIDIA Nemotron 3.5 ASR Streaming 0.6B (STT): 40 language-locales; language-ID prompt conditioning; punctuation/capitalization; released 2026-06-04.
  • Source: https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
• Infron (API Providers): Enterprise AI model inference provider; unified API; up to 35% off direct pricing; provisioned throughput plan; released 2026-06-10.
  • Source: https://infron.ai/
• Aethir Mesh (API Providers): Open-source LLM API platform; single API for open-source models (DeepSeek V4, Kimi K2.6, etc.); decentralized GPU infrastructure; released 2026-06-12.
  • Source: https://aethir.com/blog-posts/aethir-mesh-opens-the-llm-api-layer-to-everyone-leverage-top-tier-open-source-llms
• Unsloth Studio (Fine-tuning Platforms): No-code local training and inference; runs 100% offline; GGUF/Safetensors; tool-calling; web search; released 2026-06-18.
  • Source: https://unsloth.ai/
• Arkor (Fine-tuning Platforms): TypeScript framework for fine-tuning open-weight LLMs; type-safe configs; local Studio; managed GPUs; alpha 2026-06-11.
  • Source: https://github.com/WlyZhang/arkor
• Langtrain (Fine-tuning Platforms): Sovereign AI platform; fine-tune, align, deploy on own infrastructure; 20+ models; public beta 2026-06.
  • Source: https://www.langtrain.xyz/
• Dynatrace dt-evals (Evaluation & Observability): Open-source CLI for LLM/agent quality evaluation from real traces; LLM judge scoring; CI/CD integration; released 2026-06-11.
  • Source: https://www.dynatrace.com/news/blog/evaluate-llm-and-agent-quality-in-dynatrace-ai-observability/
• Currai (Evaluation & Observability): AI observability platform; prompt tracing; A/B testing; LLM evaluations; cost analytics; released 2026-06-18.
  • Source: https://www.currai.app
• Agentic CLEAR (Evaluation & Observability): IBM multi-level agent evaluation; LLM-as-judge with Key Point Analysis; ACL 2026; ...

Release v3.32: [2026-06-20] - Version 3.32 - Claude Fable 5/Mythos 5 suspended, MiniMax M3 open-weight release, Junie GA, Cursor 1.0, Amazon Bedrock AgentCore Harness GA, Databricks Genie One, Vercel eve, Omnigent, new TTS/STT models, browser automation updates, AI safety tools

Release v3.32

Release v3.32: [2026-06-20] - Version 3.32 - Claude Fable 5/Mythos 5 suspended, MiniMax M3 open-weight release, Junie GA, Cursor 1.0, Amazon Bedrock AgentCore Harness GA, Databricks Genie One, Vercel eve, Omnigent, new TTS/STT models, browser automation updates, AI safety tools

Latest Changelog Entry

[2026-06-20] - Version 3.32 - Claude Fable 5/Mythos 5 suspended, MiniMax M3 open-weight release, Junie GA, Cursor 1.0, Amazon Bedrock AgentCore Harness GA, Databricks Genie One, Vercel eve, Omnigent, new TTS/STT models, browser automation updates, AI safety tools

Added

• MiniMax M3 (Free-Source Models, Frontier Models): ~428B total / ~23B active MoE; 1M context; native multimodal (text/image/video); MiniMax Sparse Attention; $0.30/$1.20 per 1M (≤512K); open weights released ~2026-06-11; SWE-bench Verified 80.5%.
  • Source: https://huggingface.co/MiniMaxAI/MiniMax-M3, https://www.minimax.io/blog/minimax-m3
• Qwen3.7-Plus (Frontier Models): Alibaba latest Plus-tier model; 1M context; $0.40/$2.40 per 1M; released 2026-06-03.
  • Source: https://openrouter.ai/blog/announcements/may-release-spotlight/
• Step 3.7 Flash (Frontier Models): StepFun open-weight model; 1M context; free tier available; released 2026-05-28.
  • Source: https://openrouter.ai/blog/announcements/may-release-spotlight/
• Mercury 2 (Model Router Services): Inception Labs diffusion LLM; 1,009 tok/s on Blackwell; $0.25/$0.75 per 1M; 128K context; released 2026-05-12.
  • Source: https://www.inceptionlabs.ai/blog/introducing-mercury-2
• Grok Imagine Video 1.5 (Video Generation): xAI image-to-video; #1 Image-to-Video Arena (Elo 1330); native audio+video; $0.06/s; GA 2026-06-16.
  • Source: https://www.techtimes.com/articles/318635/20260618/grok-imagine-video-15-goes-live-xai-tops-ai-video-leaderboard-86-percent-below-sora.htm
• HappyHorse-1.0 (Video Generation): Alibaba ATH open-source video model; #1 Artificial Analysis leaderboard; unified Transformer; simultaneous audio/video; 15B params.
  • Source: https://www.atlascloud.ai/blog/ai-updates/happyHorse-hits-1-api-coming-soon
• Microsoft Mirage (Video Generation): Video world model with persistent spatial memory; 10.57x faster; built on Wan2.2.
  • Source: https://the-decoder.com/microsoft-researchs-mirage-gives-video-generation-a-persistent-spatial-memory-that-doesnt-forget-whats-around-the-corner/
• MAI-Voice-2 (TTS Proprietary): Microsoft expressive TTS; 15 languages; granular emotion control; custom voice from 5-60s clip; released 2026-06-02.
  • Source: https://microsoft.ai/news/mai-voice-2expressive-speech-in-10-languages/
• MAI-Transcribe-1.5 (STT): Microsoft multilingual STT; 43 languages; SOTA FLEURS; 5x faster than Gemini 3.1; keyword biasing; released 2026-06-02.
  • Source: https://microsoft.ai/news/mai-transcribe-1-5more-accurate-context-aware-and-built-for-production/
• Cohere Transcribe (STT): Open-source 2B Conformer ASR; #1 HuggingFace Open ASR Leaderboard (5.42% WER); 14 languages; Apache 2.0.
  • Source: https://cohere.com/blog/transcribe
• Higgs Audio v3 TTS (TTS Proprietary): Boson AI voice chat TTS; 100+ languages; inline emotion/style tags; zero-shot cloning.
  • Source: https://www.boson.ai/blog/higgs-audio-v3-tts
• Chatterbox Multilingual v3 (TTS Open Source): Resemble AI; 25 languages; 0.5B Llama backbone; PerTh watermarking; MIT license.
  • Source: https://www.resemble.ai/resources/chatterbox-multilingual-v3-tts-with-embedded-watermarking-for-25-languages
• MisoTTS (TTS Open Source): Miso Labs 8B emotive TTS; RVQ; conditions on text + audio; modified MIT.
  • Source: https://www.marktechpost.com/2026/06/04/miso-labs-releases-misotts-an-8b-emotive-text-to-speech-model-with-open-weights/
• MiniMax Speech 2.6 (TTS Proprietary): <250ms latency; Fluent LoRA; non-standard text handling; released 2026-06-20.
  • Source: https://www.minimax.io/news/minimax-speech-26
• Gladia Solaria-3 (STT): #1 on business audio; 5 European languages; released 2026-06-10.
  • Source: https://www.gladia.io/blog/solaria-3-speech-to-text-model-for-european-languages
• Speechmatics Melia (STT): Code-switching across 55+ languages; from $0.129/hr; released 2026-06-17.
  • Source: https://www.speechmatics.com/company/articles-and-news/introducing-melia-multilingual-speech-to-text-model
• Voxtral Realtime (STT): Mistral open-source streaming ASR; 13 languages; 480ms delay; Apache 2.0.
  • Source: https://arxiv.org/pdf/2602.11298
• Granite Embedding Multilingual R2 (Embedding Models): IBM; 97M and 311M variants; best sub-100M multilingual embedder; 32K context; Apache 2.0.
  • Source: https://huggingface.co/blog/ibm-granite/granite-embedding-multilingual-r2
• LFM2.5-Embedding-350M (Embedding Models): Liquid AI bidirectional embedder; 11 languages; runs on CPU/edge.
  • Source: https://www.liquid.ai/blog/lfm2-5-retrievers
• pplx-embed-v1 (Embedding Models): Perplexity state-of-the-art embeddings; 0.6B and 4B; MTEB Multilingual leader.
  • Source: https://research.perplexity.ai/articles/pplx-embed-state-of-the-art-embedding-models-for-web-scale-retrieval
• ML-Embed-0.6B (Embedding Models): CodeFuse AI; 3D Matryoshka Learning; ICML 2026; Apache 2.0.
  • Source: https://huggingface.co/codefuse-ai/ML-Embed-0.6B
• Sponsio (AI Safety): Runtime contract enforcement; deterministic agent safety; <0.01ms per tool call; Apache 2.0.
  • Source: https://sponsio.dev/
• ToolShield (AI Safety): ICML 2026; multi-turn safety defense; 30% attack reduction; MIT.
  • Source: https://github.com/CHATS-lab/ToolShield
• MobileMoE (SLMs): First sub-billion MoE family; 0.3B-0.9B active; 1.8-3.8x faster than dense.
  • Source: https://arxiv.org/html/2605.27358v1
• SLM-10M (SLMs): Liodon AI; leads sub-10M Open SLM Leaderboard; Apache 2.0.
  • Source: https://huggingface.co/blog/PY-AI-Dev/slm10-blog
• Vercel eve (Multi-Agent Platforms): Open-source agent framework; durable execution + sandbox + approvals; public preview 2026-06-17.
  • Source: https://vercel.com/blog/introducing-eve
• Omnigent (Multi-Agent Platforms): Databricks open-source meta-harness; compose/control/share agents; Apache 2.0.
  • Source: https://www.databricks.com/blog/introducing-omnigent-meta-harness-combine-control-and-share-your-agents
• GitKraken Kepler (Multi-Agent Platforms): Agentic Development Environment; multi-agent oversight; cross-repo coordination.
  • Source: https://www.gitkraken.com/blog/introducing-kepler-the-delivery-engine-for-agent-driven-development
• LangChain Deep Agents (Multi-Agent Platforms): Long-horizon agent harness; planning + subagents + memory; MIT.
  • Source: https://awesomeagents.ai/news/langchain-deep-agents-release/
• Cloudflare Flue (Multi-Agent Platforms): Agent framework on Cloudflare Workers; multi-agent; Durable Objects.
  • Source: https://blog.cloudflare.com/agents-platform-flue-sdk/
• Amazon Bedrock AgentCore Harness (Cloud Automation): GA 2026-06-18; 2 API calls to production agent; sandboxed runtime.
  • Source: https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-harness-is-now-generally-available-go-from-idea-to-production-grade-agent-in-minutes/
• Databricks Genie One (Cloud Automation): GA 2026-06-16; data-smart agentic coworker; iOS/Android/web; 50+ app integrations.
  • Source: https://www.databricks.com/blog/introducing-genie-one-genie-ontology-and-genie-agents
• Thoughtworks Agent/works (Cloud Automation): GA 2026-06-16; governed runtime for enterprise agents; any cloud.
  • Source: https://www.thoughtworks.com/en-us/about-us/news/2026/thoughtworks-launches-agent-works
• Kore.ai Artemis (Cloud Automation): GA 2026-05-21; AI-native agent platform; Azure launch partner.
  • Source: https://www.kore.ai/news/kore-ai-launches-artemis-the-new-generation-of-the-kore-ai-agent-platform-for-building-governing-and-optimizing-enterprise-ai
• Browserless Agent (Browser Automation): Fastest MCP browser agent; stateful sessions; command batching; released 2026-06-12.
  • Source: https://www.browserless.io/blog/introducing-browserless-agent
• Lightpanda Agent (Browser Automation): Native headless browser agent; PandaScript reproducible scripts; released 2026-06-17.
  • Source: https://lightpanda.io/blog/posts/introducing-lightpanda-agent-and-pandascript
• Webwright (Browser Automation): Microsoft SOTA on long-horizon web tasks; Codex/Claude Code plugins; released 2026-04-08.
  • Source: https://github.com/microsoft/Webwright
• SimuLang (Browser Automation): Playwright for entire desktop; browser + native apps + OS workflows; released 2026-05-19.
  • Source: https://www.simular.ai/articles/introducing-simulang-playwright-for-the-entire-desktop
• Pioneer (Fine-tuning Platforms): Fastino Labs agentic fine-tuning; synthetic dataset generation; 10-min fine-tuning.
  • Source: https://pioneer.ai/blog/introducing-pioneer
• Laminar (Evaluation & Observability): Rust-based open-source observability; ultra-fast; OpenTelemetry-native; YC S24.
  • Source: https://github.com/lmnr-ai/lmnr
• Agentic CLEAR (Evaluation & Observability): IBM Research ACL 2026; automated multi-level agent evaluation.
  • Source: https://ibm.github.io/CLEAR/
• Microsoft Foundry Toolkit for VS Code (IDE Add-ons): GA 2026-04-16; end-to-end Hosted Agent lifecycle; consolidated extension.
  • Source: https://techcommunity.microsoft.com/blog/azuredevcommunityblog/foundry-toolkit-for-vs-code-at-build-hosted-agents-end-to-end-a-smarter-toolbox-/4524941
• VS Code ACP Client (IDE Add-ons): Connects 11+ AI agents to VS Code via Agent Client Protocol; MIT.
  • Source: https://blog.brightcoding.dev/2026/06/16/stop-juggling-ai-tools-vs-code-acp-client-unifies-every-agent
• Google Agents CLI (CLI Tools): Unified ADLC for Google Cloud Agent Platform; released 2026-04-22.
  • Source: https://developers.googleblog.com/agents-cli-in-agent-platform-create-...

Release v3.31: [2026-06-18] - Version 3.31 - GLM-5.2 open-source launch, Gemma 4 12B multimodal, Antigravity 2.0 GA, Cursor Composer 2.5, GitHub Copilot usage-based billing, Google AI Ultra repricing, AIME 2026 benchmark, OpenAI GPT-Realtime-2, Stability Audio 3.0

Release v3.31

Release v3.31: [2026-06-18] - Version 3.31 - GLM-5.2 open-source launch, Gemma 4 12B multimodal, Antigravity 2.0 GA, Cursor Composer 2.5, GitHub Copilot usage-based billing, Google AI Ultra repricing, AIME 2026 benchmark, OpenAI GPT-Realtime-2, Stability Audio 3.0

Latest Changelog Entry

[2026-06-18] - Version 3.31 - GLM-5.2 open-source launch, Gemma 4 12B multimodal, Antigravity 2.0 GA, Cursor Composer 2.5, GitHub Copilot usage-based billing, Google AI Ultra repricing, AIME 2026 benchmark, OpenAI GPT-Realtime-2, Stability Audio 3.0

Added

• GLM-5.2 (Free-Source Models, Open-Source Coding Models): Zhipu AI flagship open-weight model; 753B params (MoE, 40B active), 1M context, 128K max output, MIT license; AIME 2026 99.2% (#1), Terminal-Bench 2.1 81.0, SWE-bench Pro 62.1%, FrontierSWE 74.4%; API $1.40/$4.40 per 1M; released 2026-06-13, weights open-sourced 2026-06-17.
  • Source: https://z.ai/blog/glm-5.2, https://aitoolly.com/ai-news/article/2026-06-14-zhipu-ai-releases-glm-52-a-fully-open-source-frontier-model-featuring-a-1m-context-window
• Gemma 4 12B (Free-Source Models, SLMs): Google encoder-free unified multimodal model; 12B dense, 256K context, Apache 2.0; text+image+audio input; released June 2026.
  • Source: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/
• Supertonic 3 (TTS Proprietary): Supertone on-device TTS; 99M params, 31 languages, 167x real-time on CPU; Voice Builder for custom voices; released May 2026.
  • Source: https://github.com/supertone-inc/supertonic
• Stability Audio 3.0 (TTS Open Source): Professional-grade music generation >6 min; open-weight small/medium under Apache 2.0; released May 2026.
  • Source: https://techcrunch.com/2026/05/20/stability-ai-release-a-new-audio-model-that-can-create-six-minute-songs/
• OpenAI GPT-Realtime-2 (TTS Proprietary): Speech-to-speech with GPT-5-class reasoning, tool calls, interruptions; voices Cedar and Marin; released 2026-05-07.
  • Source: https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api
• Aion 1.0 Instruct + Aion 1.0 Plan (SLMs): Microsoft on-device SLMs for Windows; Instruct is a smaller/faster SLM; Plan is a reasoning/tool-calling model for local agentic capabilities; announced Build 2026.
  • Source: https://blogs.windows.com/windowsdeveloper/2026/06/02/build-2026-furthering-windows-as-the-trusted-platform-for-development
• Microsoft Windows Agent Framework (Autonomous Agents - Local Machine): Open-sourced MIT at Build 2026; Agent Registration Service, Declarative Agent Manifest (agent.json), gRPC pub/sub bus, Memory Service; wagent CLI.
  • Source: https://blogs.windows.com/windowsdeveloper/2026/06/02/build-2026-furthering-windows-as-the-trusted-platform-for-development

Updated

• Google AI Ultra pricing (Subscription Pricing): Reduced to $200/mo (from $249.99); new $100/mo tier added (5x Pro limits, 20 TB Drive, Antigravity priority).
  • Source: https://blog.google/products-and-platforms/products/google-one/google-ai-ultra/
• Antigravity 2.0 (Web-Based IDEs, Assisted CLI Tools): GA at Google I/O 2026 on 2026-05-19; desktop app + CLI (agy) + SDK + Managed Agents API; multi-agent orchestration; successor to Gemini CLI (sunset June 18).
  • Source: https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/
• Gemini Spark (Cloud Automation): New 24/7 always-on personal AI agent; included with Google AI Ultra ($200/mo); runs on Cloud VMs; MCP connections to Canva, OpenTable, Instacart; US beta 2026-05-26.
  • Source: https://www.digitalapplied.com/blog/google-io-2026-complete-ai-announcement-guide
• Cursor Composer 2.5 (Cursor Alternatives): Launched 2026-05-18 at $0.50/$2.50 per M tokens (10x cheaper than Opus 4.7); Cursor training larger model on Colossus 2 (Q3-Q4 2026 expected).
  • Source: https://www.digitalapplied.com/blog/agentic-coding-h2-2026-what-ships-next-after-io
• GitHub Copilot (Cursor Alternatives, VS Code Specific): Usage-based billing effective June 1, 2026; $0.01/AI credit model; free tier remains 2K completions + 50 premium req/month.
  • Source: https://codex.danielvaughan.com/2026/06/05/coding-agent-landscape-june-2026-codex-cli-copilot-flex-devin-desktop-antigravity-kiro
• Android CLI 1.0 (CLI Tools): Google Android CLI stable at v1.0; enables AI agents to build Android apps from command line; "android studio" command taps Android Studio capabilities.
  • Source: https://techcrunch.com/2026/05/19/agentic-app-coding-gets-an-upgrade-with-googles-release-of-android-cli
• Arena Elo scores (Frontier Models, Benchmark Reference): Updated from arena.ai live data as of June 16, 2026: Claude Fable 5 #1 (1508), Claude Opus 4.6 Thinking #2 (1504), Claude Opus 4.7 Thinking #3 (1502), Claude Opus 4.6 #4 (1499), Claude Opus 4.7 #5 (1493), Muse Spark #6 (1487), Gemini 3.1 Pro #7 (1486), Gemini 3 Pro #8 (1486), Claude Opus 4.8 Thinking #9 (1483), GPT-5.5 High #10 (1481).
  • Source: https://arena.ai/leaderboard/text
• MCP ecosystem stats (MCP Ecosystem): Official registry 10,000+ active public servers; SDK downloads 97M+ monthly (March 2026); MCP donated to Agentic AI Foundation (Linux Foundation) Dec 2025; 2026-07-28 spec RC introduces stateless transport, Tasks primitive, enterprise auth.
  • Source: https://www.digitalapplied.com/blog/mcp-adoption-statistics-2026-model-context-protocol, https://blog.modelcontextprotocol.io/posts/2026-mcp-roadmap/
• VS Code 1.124 (IDE Add-ons): Released June 10, 2026; Agents window (preview); Autopilot enabled by default; background send for sessions; session navigation via keyboard.
  • Source: https://code.visualstudio.com/updates
• Document Version: 3.30 -> 3.31; Last Updated badge, At a Glance, Highlights bumped to 2026-06-18.

Notes

• GLM-5.2 is the second major Chinese open-source frontier model release within weeks of MiniMax M3. 753B parameters with 1M context at MIT license represents a significant accessibility milestone.
• Supertonic 3's 167x real-time on-device TTS (99M params, 31 languages) sets a new edge TTS benchmark.
• Google's Antigravity 2.0 converges developer (CLI) and consumer (Spark) agent surfaces under one harness.
• MCP 2026-07-28 spec RC introduces breaking changes: stateless transport, new required HTTP headers, Tasks lifecycle.

---

Files in Release

• docs/readme.md
• docs/readme.pdf

---

Generated on 2026-06-19T07:13:13.866Z

Release v3.29: [2026-06-17] - Version 3.29 - Fable/Mythos suspension, Kimi K2.7 Code, MiniMax M3 GA, Arena Elo refresh, stale markers

Release v3.29

Release v3.29: [2026-06-17] - Version 3.29 - Fable/Mythos suspension, Kimi K2.7 Code, MiniMax M3 GA, Arena Elo refresh, stale markers

Latest Changelog Entry

[2026-06-17] - Version 3.29 - Fable/Mythos suspension, Kimi K2.7 Code, MiniMax M3 GA, Arena Elo refresh, stale markers

Added

• Kimi K2.7 Code (Frontier Models, Free-Source Models, Coding Models, Open-Source Coding Models, Autonomous Coding Agents): Open-weight coding-focused agentic model by Moonshot AI; 1T total params, 32B active (MoE); 262K context; 30% fewer thinking tokens vs K2.6; SWE-bench Verified 78.2%, AIME 2025 91.5%; $0.95/$4.00 per 1M tokens; Modified MIT license; released 2026-06-12.
  • Source: https://huggingface.co/moonshotai/Kimi-K2.7-Code
• MiniMax M3 (Free-Source Models, Open-Source Coding Models): Open weights now released; frontier coding (SWE-Bench Pro 59.0%, SWE-Bench Verified 80.5%), 1M context, native multimodality; MIT license. Previously listed as API-only.
  • Source: https://www.minimax.io/news/minimax-m3

Updated

• Claude Fable 5 / Mythos 5 (Frontier Models, Coding Models): Marked with ⚠️ suspension notice. US government export control directive on June 12, 2026 forced Anthropic to disable global access to both models. Access may be restored for US-based users around July 1, 2026. All other Claude models unaffected.
  • Source: https://www.anthropic.com/news/fable-mythos-access
• Arena Elo scores (Frontier Models table): Updated Claude Opus 4.8 to 1483, Claude Opus 4.7 to 1502, GPT-5.5 to 1481, Gemini 3.1 Pro to 1486 based on arena.ai June 2026 data.
  • Source: https://arena.ai/leaderboard/text
• SWE-bench Verified Leaderboard: Added Kimi K2.7 Code (#21, 78.2%), MiniMax M3 (#13, 80.5%). Re-ranked entries #13-#26 accordingly.
  • Source: https://benchlm.ai/benchmarks/sweVerified
• AIME 2025 Leaderboard: Updated with BenchLM.ai June 2026 data. MAI-Thinking-1 now #1 (97%), Kimi K2.5 variants at #2 (96.1%).
  • Source: https://benchlm.ai/benchmarks/aime2025
• Top Models by Category: Coding #1 changed to Claude Opus 4.8 (Fable 5 suspended); Agentic Performance #1 changed to Claude Opus 4.8; Open Source #1 changed to MiniMax M3.
• Codex CLI (Autonomous Coding Agents): Updated description to reflect Rust-native rewrite, v0.140+, /goal persistent workflows, Chrome extension support.
  • Source: https://github.com/openai/codex
• Gemini CLI deprecation note (Assisted CLI Tools): Google confirmed shutdown date of June 18, 2026; migrate to Antigravity CLI (agy).
  • Source: https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/

Removed / Deprecated

• Stale ⭐ markers removed: Removed ⭐ from entries older than 30 days from 2026-06-17: Gemini 3.5 Flash, Gemini Omni Flash, Qwen3.7-Max, Qwen3.7-Plus-Preview, Grok 4 Fast, Grok 4.20, MiniMax M3 (re-added as still within 30 days of GA), NVIDIA Nemotron 3 Ultra, MAI-Thinking-1, MAI-Code-1-Flash, Kimi Code CLI. Retained ⭐ on: MiniMax M3 (2026-06-01, still within 30 days), Kimi K2.7 Code (2026-06-12), Claude Fable 5 (2026-06-09), Claude Mythos 5 (2026-06-09).
• Document Version: 3.28 -> 3.29; Last Updated badge updated to 2026-06-17.

---

Files in Release

• docs/readme.md
• docs/readme.pdf

---

Generated on 2026-06-18T06:44:50.318Z

Release v3.26: [2026-06-08] - Version 3.26 - Nemotron 3 Ultra, Colab CLI, OpenRouter Speech API, benchmark & staleness fixes

Release vv3.26

Release v3.26: [2026-06-08] - Version 3.26 - Nemotron 3 Ultra, Colab CLI, OpenRouter Speech API, benchmark & staleness fixes

Latest Changelog Entry

[2026-06-08] - Version 3.26 - Nemotron 3 Ultra, Colab CLI, OpenRouter Speech API, benchmark & staleness fixes

Added

• NVIDIA Nemotron 3 Ultra (Frontier Models + Free-Source Models): 550B/55B active MoE; 262K context (BF16) / 1M (NVFP4 Blackwell); open weight; launched at Computex June 2026; orchestration-focused, Mamba+Transformer hybrid.
  • Source: https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/
• Google Colab CLI (Assisted CLI Tools): Connect local terminal to remote Colab GPU/TPU runtimes; AI agent compatible; released 2026-06-06.
  • Source: https://www.marktechpost.com/2026/06/06/googles-new-colab-cli-lets-developers-and-ai-agents-run-python-on-remote-colab-gpus-and-tpus-from-the-terminal/

Updated

• Kimi K2.6 (Frontier Models, Output Limits): Context window corrected from 256K to 262K tokens (262,144); max output updated to 262K.
  • Source: https://huggingface.co/moonshotai/Kimi-K2.6
• GPT-5.5 (Frontier Models, Benchmark Reference): Added Arena Elo 1495, SWE-bench Verified 88.5%, AIME 2025 99.9% from leaderboard data; corrected GPQA Diamond to 93.6%.
  • Source: https://benchlm.ai/benchmarks/sweVerified
• OpenRouter (Unified APIs): Updated model count to 370+; added Speech & Transcription APIs, Model Fusion, private models, enterprise workspace controls (June 4, 2026).
  • Source: https://openrouter.ai/announcements/all
• Microsoft Agent Framework (Multi-Agent Platforms): Renamed from AutoGen 1.0 to reflect merger with Semantic Kernel; v1.0 GA April 2026.
  • Source: https://uvik.net/blog/agentic-ai-frameworks/
• Stale ⭐ removed: GPT-5.5 Instant (2026-05-05), Grok 4.3 (2026-05-06), Mistral Medium 3.5 (2026-04-29) — all older than 30 days from 2026-06-08.

---

Files in Release

• docs/readme.md
• docs/readme.pdf

---

Generated on 2026-06-07T23:01:20.917Z

Release v3.25: [2026-06-07] - Version 3.25 - IDE Add-ons, Output Specs, Benchmark Reference updates

Release vv3.25

Release v3.25: [2026-06-07] - Version 3.25 - IDE Add-ons, Output Specs, Benchmark Reference updates

Latest Changelog Entry

[2026-06-07] - Version 3.25 - IDE Add-ons, Output Specs, Benchmark Reference updates

Added

• JetBrains AI Assistant for VS Code (IDE Add-ons / VS Code Specific): Public preview; multi-file edits, Mellum LLM; JetBrains AI in VS Code/Cursor.
  • Source: https://www.jetbrains.com/aia-vscode/
• ReSharper for VS Code (IDE Add-ons / VS Code Specific): C# code analysis + refactoring inside VS Code/Cursor; released 2026-03-05.
  • Source: https://blog.jetbrains.com/dotnet/2026/03/05/resharper-for-visual-studio-code-cursor-and-compatible-editors-is-out/
• Mistral Medium 3.5 (Output Token Limits + Benchmark Reference): Added spec rows for 256K context, 8K output, $1.50/$7.50.

Updated

• GitHub Copilot Agent Mode (IDE Add-ons): Added note that it auto-opens PRs with self-review.

---

Files in Release

• docs/readme.md
• docs/readme.pdf

---

Generated on 2026-06-07T19:58:57.745Z

Release v3.15: [2026-05-28] - Version 3.15 - Claude Opus 4.8, Grok pricing cuts, GPU Cloud pricing, model data updates

Release vv3.15

Release v3.15: [2026-05-28] - Version 3.15 - Claude Opus 4.8, Grok pricing cuts, GPU Cloud pricing, model data updates

Latest Changelog Entry

[2026-05-28] - Version 3.15 - Claude Opus 4.8, Grok pricing cuts, GPU Cloud pricing, model data updates

Added

• Claude Opus 4.8: New entry across all relevant tables — Frontier Models, Cached & Batch Pricing, Benchmark Reference, SWE-bench Verified Leaderboard, Commercial Coding Models, and Cost Analysis. Released 2026-05-28 by Anthropic; $5.00/$25.00 per 1M tokens (same as Opus 4.7), fast mode $10/$50 (3x cheaper than Opus 4.7 fast mode). SWE-bench Verified 88.6%, SWE-bench Pro 69.2%. Added to Sort by Latest Update comparison table and Primary Sources.
• GPU Clouds pricing table: Replaced the bare provider listing with detailed A100 80GB and H100 80GB hourly pricing for RunPod, Vast.ai, Lambda Labs, DigitalOcean, Vultr, Hyperbolic, Cerebrium, Modal Labs, Replicate, and other providers.

Changed

• Grok 4.20 pricing: Updated from $2.00/$6.00 to $1.25/$2.50 across Frontier Models, Cached & Batch Pricing, Regional Availability, Output Token Limits, and Cost Analysis tables (83% output price cut per xAI developer docs).
• Grok 4 Fast pricing: Output price updated from $1.50 to $0.50 (aliased to Grok 4.3 per xAI API). Updated across Frontier Models, Cached & Batch Pricing, Output Token Limits, Commercial Coding Models, and Sort by Price tables. Note: Grok 4 Fast is now an alias for Grok 4.3 with unified pricing.
• xAI price cuts effective 2026-05-28: Grok 4.20 (reasoning, non-reasoning, multi-agent), Grok 4.3, and Grok Build all now at $1.25/$2.50 per 1M tokens; cached input $0.20/1M.
• Frontier Models Top Table: Updated Coding #1 to Claude Opus 4.8; Agentic Performance #1 to Claude Opus 4.8; Free & Budget reordered with cheapest first (Gemini 3.1 Flash-Lite $0.25, Grok 4.3 $1.25, Claude Haiku 4.5 $1.00).
• Model Labs (Direct): Added GPT-5.5, GPT-5.5 Instant, GPT-5.5 Pro to OpenAI row; added Claude Opus 4.8 to Anthropic row.
• Google Gemini API free tier: Updated references from Gemini 2.5 Pro/Flash/Flash-Lite to Gemini 3.1 Pro/3 Flash/3.1 Flash-Lite in Free AI APIs for Coding section.
• Project Mariner pricing: Updated Google AI Ultra reference from $249.99 to $100/mo (5× limits tier added at I/O 2026).
• Cost Analysis Pricing Tiers: Added Grok 4.3 to Mid-range tier ($0.60–$15.00/1M); updated Premium tier to include GPT-5.5 Pro and Claude Opus 4.8/4.7; added note about Grok 4 Fast aliasing to 4.3 in Budget tier.
• Sort by Latest Update: Added Claude Opus 4.8 entry; added "Grok 4.20 / 4.3 / 4 Fast" row noting major price cuts.
• Release Windows table: Updated Grok entry to note price cuts; expanded Claude Opus 4.8 entry with SWE-bench and fast mode details.
• Document metadata: Version 3.14 → 3.15; no change to Last Updated (already 2026-05-28).

Fixed

• Grok 4.20 context note: Updated from "Largest context window among Western closed models" to note all variants at $1.25/$2.50 pricing.
• Grok 4 Fast context: Updated from "2M context" note to "128K context; aliased to Grok 4.3; output $0.50/1M" across all relevant tables.

Sources (verification, 2026-05-28 UTC)

• Claude Opus 4.8 announcement — anthropic.com/news/claude-opus-4-8 (released 2026-05-28, $5/$25, 88.6% SWE-bench, 69.2% SWE-bench Pro)
• xAI Developer Models — docs.x.ai/developers/models (Grok 4.3 $1.25/$2.50; Grok 4.20 variants $1.25/$2.50; Grok 4 Fast → Grok 4.3 alias)
• Grok 4 Fast pricing — docs.x.ai/developers/models/grok-4-fast (confirms alias to Grok 4.3)
• DeepSeek API Pricing — api-docs.deepseek.com/quick_start/pricing (75% discount V4-Pro promo until 2026-05-31; V4-Flash permanent low rates)
• RunPod, Vast.ai, Lambda Labs, DigitalOcean, Vultr, Hyperbolic, Cerebrium, Modal Labs, Reify, Together AI pricing pages (May 2026)

Changed

• Document metadata: docs/readme.md document version 3.13 → 3.14; Last Updated set to 2026-05-28 00:00 UTC.
• Pricing format standardization: Corrected $X / $X format to $X.XX / $X.XX per CLAUDE.md formatting standards for:
  • Gemini 3.1 Pro ($2 / $12 → $2.00 / $12.00)
  • Claude Mythos Preview ($25 / $125 → $25.00 / $125.00)
• ⭐ flag corrections: Removed incorrect ⭐ flag from Claude Mythos Preview (2026-04-07 is more than 30 days before 2026-05-28).

Files in Release

• docs/readme.md
• docs/readme.pdf

---

Generated on 2026-05-29T10:54:34.249Z

Version 3.0 Major Update

🚀 Version 3.0 Major Update

Added

• Browser Automation Enhancements: Added API Access, Multi-Agent, and Parallel Sessions columns to Standalone AI Browsers table
• New Browser: Sidekick Browser (Windows, macOS, Linux) with AI assistant and natural language tab management
• New Developer Libraries: AgentQL (natural language web querying), ScrapeGraphAI (natural language scraping), WebVoyager (autonomous browsing research)
• New Local Agent: Khoj (supports Ollama, LM Studio, OpenAI API)
• New Cloud Agents: Perplexity Computer (Pro /mo), OpenAI Computer Use (API: /M input, /M output)
• Multi-Agent & Parallel Execution Summary: New dedicated section categorizing tools by parallel execution capability
• Pricing Column: Added to Developer Libraries table with detailed pricing

Changed

• Document Version: 2.9 → 3.0
• SWE-bench Rankings: Updated with GPT-5.5 Pro (#1, 92.3%), GPT-5.5 (#2, 88.5%), Claude Opus 4.7 (#3, 87.6%)
• Reasoning Benchmarks: Updated with GPT-5.5 Pro (#1, 100% AIME, 78.5% ARC-AGI-2)
• Model Updates: Kimi K2.5 → Kimi K2.6, DeepSeek-V3.1 → DeepSeek-V3.2
• IDE Updates: Cursor 3.1 → 3.2, Windsurf 1.9552+ → 2.0.0
• Claude Code: Updated to 2.2.1+ (added multi-session support, Opus 4.7 support)
• Commercial Coding Models: Added GPT-5.5 Pro (.00/.00 per 1M, 92.3% SWE-bench)

Sources

• Official provider documentation (OpenAI, Anthropic, Microsoft, Perplexity)
• SWE-bench Verified benchmark results (May 2026)
• AIME 2025 and ARC-AGI-2 benchmark data

---

Full Changelog: https://github.com/Shaerif/AI_Models_Matrix/blob/main/CHANGELOG.md

Awesome List Compliance Update

Changes

• GitHub Actions workflows added (link-check, lint, awesome-lint)
• 7 model entries flagged as unverified
• GPT-5.3-Codex pricing corrected
• CHANGELOG.md, CODE_OF_CONDUCT.md, SECURITY.md added
• CONTRIBUTING.md expanded

v2.1 - Model Specifications & Awesome-List Compliance

What's New in v2.1

Model Specifications Section

• Output token limits for all frontier models
• Cached & batch pricing columns
• Speed & latency metrics (tokens/sec, TTFT)
• Training data cutoff dates
• Multilingual support details
• Structured output & function calling support
• Regional availability info

Awesome-List Compliance

• Repository logo added
• TOC flattened to max 1 level of nesting
• TOC renamed from 'Table of Contents' to 'Contents'
• License section removed (badge + LICENSE file sufficient)

Data Quality

• Fixed model naming inconsistencies across all tables
• Removed Perplexity pricing link (data rows remain)
• Cleaned up all cross-references

Previous (v2.0)

• 10 new AI Infrastructure sections (Embedding, Video Gen, Speech/TTS, Safety, RAG, Fine-tuning, Eval/Observability, MCP, Routers, SLMs)
• Comprehensive Benchmark Reference (12 benchmarks x 24 models)
• Real benchmark scores replaced all subjective labels
• GitHub Discussions enabled
• CONTRIBUTING.md, issue templates, repo metadata