{
  "$schema_version": 1,
  "generated_at": "2026-06-16",
  "note": "Reverse-chronological feed of changes to the model lineup. Append-only. The weekly freshness sweep adds entries here; humans can hand-add for clarifications.",
  "entries": [
    {
      "id": "2026-06-15-moonshot-2",
      "date": "2026-06-15",
      "vendor": "moonshot",
      "title": "Kimi K2.7-Code HighSpeed (Beta)",
      "summary": "HighSpeed variant rolling out to Kimi Code Beta — ~180 tok/s median, up to ~260 tok/s on short context. Roughly 6× the standard K2.7-Code throughput.",
      "type": "new_model",
      "source_url": "https://www.techtimes.com/articles/318414/20260615/kimi-k27-code-adds-highspeed-mode-skips-independent-benchmark-submission.htm"
    },
    {
      "id": "2026-06-15-xai-1",
      "date": "2026-06-15",
      "vendor": "xai",
      "title": "Grok V9-Medium expected mid-June",
      "summary": "Musk announced 2026-05-25 that V9-Medium (1.5T parameters, 3× current production size) finished training. Public release expected mid-June 2026.",
      "type": "upcoming",
      "source_url": "https://x.ai/news"
    },
    {
      "id": "2026-06-13-zai-1",
      "date": "2026-06-13",
      "vendor": "zai",
      "title": "GLM-5.2 — 1M-context coding-first frontier",
      "summary": "Live across every GLM Coding Plan tier. 1M usable context, coding-first positioning. Compatible with Claude Code, Cline, OpenCode, Roo Code, Goose, Crush, OpenClaw, Kilo Code. MIT-licensed weights coming.",
      "type": "new_model",
      "source_url": "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"
    },
    {
      "id": "2026-06-12-moonshot-1",
      "date": "2026-06-12",
      "vendor": "moonshot",
      "title": "Kimi K2.7-Code — 1T MoE open-weight coder",
      "summary": "1T MoE (32B active, 384 experts), 256K context, Modified MIT license. Cuts reasoning tokens 30% vs K2.6; +21.8% on Kimi Code Bench v2. Available on Hugging Face, Kimi API, Kimi Code CLI.",
      "type": "new_model",
      "source_url": "https://cryptobriefing.com/kimi-k2-7-code-open-source-release/"
    },
    {
      "id": "2026-06-09-anthropic-1",
      "date": "2026-06-09",
      "vendor": "claude",
      "title": "Claude Fable 5 + Mythos 5 — new Mythos-class tier",
      "summary": "First models in the new Mythos-class tier, sitting above Opus. Fable 5 brings Mythos-class capability to general use with safeguards; Mythos 5 remains restricted to approved programs and trusted-access use.",
      "type": "new_model",
      "source_url": "https://www.anthropic.com/news"
    },
    {
      "id": "2026-06-09-moonshot-1",
      "date": "2026-06-09",
      "vendor": "moonshot",
      "title": "Kimi Work — desktop app, 300 sub-agents local",
      "summary": "Desktop application powered by Kimi K2.6 that coordinates up to 300 specialized sub-agents on the user's local machine — no cloud required.",
      "type": "new_model",
      "source_url": "https://cryptobriefing.com/moonshot-ai-kimi-work-300-agents-desktop/"
    },
    {
      "id": "2026-06-08-google-1",
      "date": "2026-06-08",
      "vendor": "google",
      "title": "Gemini 3.5 Flash GA in Gemini Enterprise",
      "summary": "3.5 Flash became generally available in Gemini Enterprise, enabled by default starting 2026-06-08. Frontier-intelligence Flash tier — fast, cheap, multimodal.",
      "type": "new_model",
      "source_url": "https://docs.cloud.google.com/gemini/enterprise/docs/release-notes"
    },
    {
      "id": "2026-06-04-blackforest-1",
      "date": "2026-06-04",
      "vendor": "blackforest",
      "title": "FLUX.2 [klein] — first FLUX on consumer hardware",
      "summary": "Partnership with ASUS + NVIDIA. First FLUX model optimized to run on consumer hardware — shipping with ASUS ProArt laptops.",
      "type": "new_model",
      "source_url": "https://blogs.nvidia.com/blog/rtx-ai-garage-flux-2-comfyui/"
    },
    {
      "id": "2026-06-03-openai-1",
      "date": "2026-06-03",
      "vendor": "openai",
      "title": "GPT-Rosalind life-sciences expansion",
      "summary": "Capability expansion for vetted developers + U.S. government partners — biological reasoning, medicinal chemistry, genomics analysis, experimental workflow capabilities. Restricted/trusted-access tier.",
      "type": "capability",
      "source_url": "https://openai.com/research/index/release/"
    },
    {
      "id": "2026-06-01-alibaba-1",
      "date": "2026-06-01",
      "vendor": "alibaba",
      "title": "Qwen 3.7 Plus — low-cost multimodal agent",
      "summary": "Generally available 2026-06-01. Multimodal (image + video). 1M-token context. Roughly 1/6 the per-token price of the text-only Qwen 3.7 Max.",
      "type": "new_model",
      "source_url": "https://www.marktechpost.com/2026/06/02/alibabas-qwen-team-launches-qwen3-7-plus-adding-vision-deep-reasoning-tool-invocation-and-autonomous-iteration-on-the-bailian-platform/"
    },
    {
      "id": "2026-06-01-xai-1",
      "date": "2026-06-01",
      "vendor": "xai",
      "title": "Grok Build 0.1 — agentic coding model (public beta)",
      "summary": "Purpose-built coding model trained for agentic workflows. Text + image input, 256K-token context. Available in public beta on the xAI API.",
      "type": "new_model",
      "source_url": "https://x.ai/news"
    },
    {
      "id": "2026-05-29-openai-1",
      "date": "2026-05-29",
      "vendor": "openai",
      "title": "GPT-Rosalind launched (Biodefense)",
      "summary": "OpenAI launched Rosalind Biodefense — expanded trusted access to GPT-Rosalind for vetted developers and U.S. government partners advancing biodefense, public health, and pandemic preparedness.",
      "type": "new_model",
      "source_url": "https://openai.com/research/index/release/"
    },
    {
      "id": "2026-05-28-blackforest-1",
      "date": "2026-05-28",
      "vendor": "blackforest",
      "title": "FLUX Virtual Try-On",
      "summary": "Garment try-on for retail — sub-4-second generations across thousands of products. High garment consistency.",
      "type": "new_model",
      "source_url": "https://bfl.ai/blog"
    },
    {
      "id": "2026-05-28-anthropic-1",
      "date": "2026-05-28",
      "vendor": "claude",
      "title": "Claude Opus 4.8 + Dynamic Workflows preview",
      "summary": "Opus 4.8 shifts emphasis to honesty and reliability — ~4× less likely than 4.7 to let flaws in code it has written pass unremarked. Ships alongside a Dynamic Workflows research preview in Claude Code that orchestrates hundreds of parallel subagents.",
      "type": "new_model",
      "source_url": "https://www.anthropic.com/news"
    },
    {
      "id": "2026-05-21-blackforest-1",
      "date": "2026-05-21",
      "vendor": "blackforest",
      "title": "FLUX Erase",
      "summary": "Removes masked objects, shadows, and reflections — reconstructs the scene behind them.",
      "type": "new_model",
      "source_url": "https://bfl.ai/blog"
    },
    {
      "id": "2026-05-20-alibaba-1",
      "date": "2026-05-20",
      "vendor": "alibaba",
      "title": "Qwen 3.7 Max — Agent Frontier flagship",
      "summary": "Unveiled at Alibaba Cloud Summit Hangzhou. 1M-token context, 56.6 on Artificial Analysis Intelligence Index. Tops Claude Opus 4.6 Max on Terminal-Bench 2.0, SWE-Bench Pro, MCP-Atlas. Closed weights.",
      "type": "new_model",
      "source_url": "https://www.scmp.com/tech/big-tech/article/3354212/alibaba-unveils-new-qwen-model-custom-chips-bid-become-chinas-ai-factory"
    },
    {
      "id": "2026-05-19-google-1",
      "date": "2026-05-19",
      "vendor": "google",
      "title": "Gemini 3.5 Pro announced at Google I/O 2026",
      "summary": "Frontier multimodal flagship announced at I/O 2026 with CEO Sundar Pichai promising general availability 'next month'. Still in limited preview as of 2026-06-16; GA expected late June.",
      "type": "new_model",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog"
    },
    {
      "id": "2026-05-14-blackforest-1",
      "date": "2026-05-14",
      "vendor": "blackforest",
      "title": "FLUX.1 Tools — control + steerability suite",
      "summary": "Suite designed to add control and steerability to the base text-to-image FLUX.1 model (depth, canny, fill, redux).",
      "type": "new_model",
      "source_url": "https://bfl.ai/blog"
    },
    {
      "id": "2026-05-07-openai-1",
      "date": "2026-05-07",
      "vendor": "openai",
      "title": "New realtime voice models in the OpenAI API",
      "summary": "Realtime voice models that can reason, translate, and transcribe speech in a single low-latency pipeline.",
      "type": "new_model",
      "source_url": "https://openai.com/research/index/release/"
    },
    {
      "id": "2026-05-05-openai-1",
      "date": "2026-05-05",
      "vendor": "openai",
      "title": "GPT-5.5 Instant — new ChatGPT default",
      "summary": "ChatGPT's default model updated to GPT-5.5 Instant. Smarter, more accurate answers, reduced hallucinations, improved personalization controls.",
      "type": "new_model",
      "source_url": "https://techcrunch.com/2026/05/05/openai-releases-gpt-5-5-instant-a-new-default-model-for-chatgpt/"
    },
    {
      "id": "2026-05-10-shell-1",
      "date": "2026-05-10",
      "vendor": "shell",
      "title": "Repository data layer introduced",
      "summary": "Volatile facts (model IDs, prices, dates, status) extracted to data/models.json. Manuals render lineup tables from JSON. Weekly freshness sweep will append entries here automatically.",
      "type": "infra",
      "source_url": null
    },
    {
      "id": "2026-04-30-xai-1",
      "date": "2026-04-30",
      "vendor": "xai",
      "title": "Grok 4.3 — current flagship",
      "summary": "Released 2026-04-30. $1.25/$2.50 — cheapest hosted frontier on list price. 1M context, native video input, always-on reasoning. ~40% input price cut vs Grok 4.",
      "type": "new_model",
      "source_url": "https://docs.x.ai/docs/release-notes"
    },
    {
      "id": "2026-04-24-deepseek-1",
      "date": "2026-04-24",
      "vendor": "deepseek",
      "title": "DeepSeek V4 (Pro + Flash) launched",
      "summary": "V4 Pro: 1.6T / 49B-active MoE, 1M context, MIT-licensed. Open-source SOTA on agentic-coding benchmarks. V4 Flash: 284B / 13B-active, same context. 75% Pro discount runs through 2026-05-31.",
      "type": "new_model",
      "source_url": "https://api-docs.deepseek.com/news/news260424"
    },
    {
      "id": "2026-04-23-openai-1",
      "date": "2026-04-23",
      "vendor": "openai",
      "title": "GPT-5.5 / 5.5 Pro in ChatGPT (API TBA)",
      "summary": "Available in ChatGPT (Plus/Pro/Business/Enterprise) + Codex now. Built around long-running goal completion. API pricing TBA.",
      "type": "new_model",
      "source_url": "https://help.openai.com/en/articles/9624314-model-release-notes"
    },
    {
      "id": "2026-04-20-moonshot-1",
      "date": "2026-04-20",
      "vendor": "moonshot",
      "title": "Kimi K2.6 — 1T-MoE open-weight agent flagship",
      "summary": "Modified MIT license. 1T total / 32B active. 262K context, 16K max output. Agent Swarm: 300 sub-agents × 4000 steps per run. $0.60/$2.50 on Moonshot API.",
      "type": "new_model",
      "source_url": "https://huggingface.co/moonshotai"
    },
    {
      "id": "2026-04-16-claude-1",
      "date": "2026-04-16",
      "vendor": "claude",
      "title": "Claude Opus 4.7 — current flagship",
      "summary": "$5/$25 per M tokens. 200K context. New xhigh effort dial. Major vision lift (98.5% visual-acuity). Task budgets (public beta), file-system memory. Tokenizer change: 1.0–1.35× more tokens than 4.6.",
      "type": "new_model",
      "source_url": "https://www.anthropic.com/news"
    },
    {
      "id": "2026-04-08-zai-1",
      "date": "2026-04-08",
      "vendor": "zai",
      "title": "GLM-5.1 open-sourced",
      "summary": "745B / 44B-active MoE, MIT license, 200K context. DeepSeek Sparse Attention integrated. From China's first publicly-traded AI lab. Subscription preview late March.",
      "type": "new_model",
      "source_url": "https://huggingface.co/zai-org"
    },
    {
      "id": "2026-04-alibaba-1",
      "date": "2026-04-15",
      "vendor": "alibaba",
      "title": "Qwen 3.6 wave — Max + Plus + multimodal",
      "summary": "Qwen 3.6 Max (proprietary flagship, 1M+ context, agentic + visual reasoning). Qwen 3.6 Plus (high-perf, speed/cost tier). Wan 2.7 (video gen) + HappyHorse 1.0 (image-to-video, top-ranked) added to Model Studio.",
      "type": "new_model",
      "source_url": "https://modelstudio.console.alibabacloud.com"
    },
    {
      "id": "2026-03-31-xai-2",
      "date": "2026-03-31",
      "vendor": "xai",
      "title": "Grok 4.20 — 2M-context fast frontier",
      "summary": "$2.00/$6.00 per M tokens. 2M context — largest in the field alongside Gemini 3.1 Pro. Strong agentic tool calling.",
      "type": "new_model",
      "source_url": "https://docs.x.ai/docs/release-notes"
    },
    {
      "id": "2026-03-05-openai-1",
      "date": "2026-03-05",
      "vendor": "openai",
      "title": "GPT-5.4 + Computer Use SOTA",
      "summary": "75.0% on OSWorld-Verified (vs 47.3% for GPT-5.2). Beats human baseline 72.4%. 95% first-try / 100% within 3 tries on real portals.",
      "type": "new_model",
      "source_url": "https://help.openai.com/en/articles/9624314-model-release-notes"
    },
    {
      "id": "2026-03-05-lightricks-1",
      "date": "2026-03-05",
      "vendor": "specvideo",
      "title": "Lightricks LTX 2.3 — open-source 4K video",
      "summary": "22B Diffusion Transformer. Native 4K @ 50fps with synchronized audio. FP8 quantized fits 24GB VRAM (RTX 4090/5090). ~$0.04/sec hosted.",
      "type": "new_model",
      "source_url": "https://huggingface.co/Lightricks/LTX-2.3"
    },
    {
      "id": "2026-02-19-google-1",
      "date": "2026-02-19",
      "vendor": "google",
      "title": "Gemini 3.1 Pro — 2M context, ARC-AGI-2 SOTA",
      "summary": "$2/$12 per M (≤200K input); $4/$18 above. 77.1% on ARC-AGI-2 (vs 31.1% for Gemini 3 Pro). 94.3% on GPQA Diamond. Deep Think mode.",
      "type": "new_model",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog"
    },
    {
      "id": "2026-02-12-bytedance-1",
      "date": "2026-02-12",
      "vendor": "bytedance",
      "title": "Seedance 2.0 — unified audio-video",
      "summary": "Accepts up to 9 image / 3 video / 3 audio references per prompt. 4-15s multi-shot output with dual-channel audio. Most flexible reference inputs in video gen.",
      "type": "new_model",
      "source_url": "https://seed.bytedance.com/en"
    },
    {
      "id": "2026-02-11-zai-1",
      "date": "2026-02-11",
      "vendor": "zai",
      "title": "GLM-5 — first 745B MoE flagship",
      "summary": "745B total / 40-44B active MoE. 200K context. Pre-training data 28.5T tokens (up from 23T on GLM-4.5).",
      "type": "new_model",
      "source_url": "https://huggingface.co/zai-org"
    },
    {
      "id": "2026-02-alibaba-1",
      "date": "2026-02-08",
      "vendor": "alibaba",
      "title": "Qwen 3.5 Series — 60% cheaper, 8× faster",
      "summary": "Open-source + API. Native multimodal (text + image + video). Includes Qwen 3.5 Omni for full audio + multimodal.",
      "type": "price_change",
      "source_url": "https://huggingface.co/Qwen"
    }
  ]
}