⏰ OpenRouter Limited-Time Free Models

Several premium models remain available with free tiers through late July and August:

  • Tencent: Hy3 (free) — expires July 21 (1 day left), 262K context
  • Poolside: Laguna M.1 — free tier available through July 28, 262K context; paid plan at $0.20/M tokens input, $0.40/M tokens output
  • OpenAI: GPT-5.3 Chat — free trial until August 10, 128K context ($1.75/M tokens input, $14/M tokens output)
  • OpenAI: GPT-5.2 Chat — free trial until August 10, 128K context ($1.75/M tokens input, $14/M tokens output)

🆕 New on OpenRouter

Thinking Machines: Inkling — a new long-context model with 1,048,576 tokens of context, priced at $1.00/M tokens input and $4.05/M tokens output. Designed for deep reasoning tasks over large documents.

MoonshotAI: Kimi K3 — Moonshot’s latest flagship arrives on OpenRouter with 1,048,576 context at $3.00/M tokens input and $15.00/M tokens output. Targets complex analysis and long-document understanding.

Meta: Muse Spark 1.1 — an updated version of Meta’s Muse Spark model, offering 1,048,576 context at $1.25/M tokens input and $4.25/M tokens output. A solid mid-range option for production workloads.

OpenRouter: Auto Router (Beta) — OpenRouter launched an intelligent auto-routing system that selects the best model per request, supporting up to 2,000,000 tokens of context.

🤗 Trending on Hugging Face

New uploads to the Hub include several fine-tuned LoRAs and adapters. Notable standouts:

⭐ GitHub Trending: AI Edition

nethical6/conversation-steganography ⭐767 — Use LLMs to hide messages inside normal-looking conversations. A creative Go-based project demonstrating a novel application of language models for covert communication.

Blueturboguy07/cue ⭐569 — An open-source macOS AI copilot that floats over your screen, watches meetings, and automatically hides from screen shares. A privacy-first ambient assistant built with JavaScript.

PengZhang64/circuit-framework ⭐484 — A multi-agent LLM trading research system in Python. The Circuit Framework coordinates multiple AI agents for algorithmic trading strategies.

v-modal/vmodal_sdk_flutter ⭐396 — V-Modal AI: “Search anything anywhere” SDK for Flutter, bringing multi-platform semantic search to mobile apps.

MIgHTy-alIeN/Trading-Bot ⭐350 — An arbitrage bot smart contract connected to an external automation script. Built with Solidity for DeFi trading.

📰 AI News Roundup

Anthropic “Honeycomb” Model Leaked in Cursor

A model codenamed “Honeycomb” reportedly appeared inside Cursor’s configuration files, fueling speculation that it could be Claude Opus 5, Anthropic’s next flagship model. The leak is not an official announcement, and details beyond the model name remain unconfirmed. (Source: Times of AI, July 17)

Google Delays Gemini 3.5 Pro

Google has halted the public release of Gemini 3.5 Pro after internal testing revealed coding capabilities fell short of standards. The model, revealed at Google I/O in May, was expected to launch in June. The delay comes as OpenAI, Meta, Anthropic, and xAI prepare new coding-focused models. (Source: Times of AI, July 17)

EU Pushes Google to Open Android to Rival AI Assistants

The European Union is asking Google to open key Android features and data to competitors under the Digital Markets Act, giving OpenAI and others greater access to services previously dominated by the tech giant. Google warns the changes could weaken user privacy guardrails. (Source: Times of AI, July 17)

OpenAI Targets 2027 for First Hardware: ChatGPT Speaker

OpenAI’s first major hardware device will be a screenless smart speaker running GPT-Live, according to a Bloomberg report. The device targets a 2027 launch and represents OpenAI’s first foray into consumer hardware. (Source: AI Daily Post)

OpenAI’s Miles Wang in Talks for $2B AI Drug Discovery Startup

Miles Wang is leaving OpenAI to start a company applying AI models to drug discovery, reportedly valued at $2 billion. The move signals accelerating crossover between frontier AI labs and biotech. (Source: AI Daily Post, July 15)

Mistral Vibe for Code Leads Multi-Agent Programming Benchmark

Mistral Vibe for Code tops a new multi-agent programming benchmark where coding agents take feature requests straight to pull requests without human intervention. (Source: AI Daily Post, July 15)

Indonesia is moving to become the first Southeast Asian nation to define AI’s role in copyright law. The draft legislation keeps copyright with humans while allowing AI-assisted creation — a potential template for the region. (Source: Times of AI, July 17)

Publishers Sue Google Over Unauthorized AI Book Training

Hachette, Cengage, Elsevier, and novelist Scott Turow have joined a class action against Google, alleging unauthorized use of copyrighted books for AI training. (Source: AI Daily Post, July 14)

PrismML’s Bonsai 27B Runs Qwen3.6 on Laptops

PrismML’s Bonsai 27B uses 1-bit and ternary quantization to run Qwen3.6-class models on consumer laptops, pushing the frontier of on-device inference. (Source: AI Daily Post)

Anthropic Launches Claude for Teachers

Anthropic introduced Claude for Teachers, pledging not to train on student data — a move aimed at the education market where data privacy is paramount. (Source: AI Daily Post)

DeepSeek Seeks More Capital After $7B Round

DeepSeek is already seeking additional capital just weeks after closing a reported $7 billion funding round, signaling aggressive expansion plans. (Source: AI Daily Post)

💡 Key Trends

  1. Model Release Slowdown, Not an AI Winter — After a blistering first half of 2026, July has brought several delays (Gemini 3.5 Pro) and leaks (Honeycomb/Opus 5) rather than official launches. Labs appear to be raising quality bars rather than slowing R&D.

  2. Regulatory Wave Intensifies — From the EU’s Digital Markets Act targeting Android AI assistant monopolies to Indonesia’s copyright framework and publishers’ class action against Google, AI regulation is accelerating globally across multiple dimensions simultaneously.

  3. AI Hardware Takes Shape — OpenAI’s screenless ChatGPT speaker and PrismML’s on-device Bonsai 27B represent opposite ends of the hardware spectrum: cloud-connected consumer devices vs. local laptop inference. Both signal the AI industry maturing beyond pure software.

  4. Agentic Coding Becomes Real — Mistral Vibe for Code’s multi-agent benchmark results show that coding agents can now handle end-to-end feature requests without human intervention, marking a tangible milestone in autonomous software development.

📊 Today’s Snapshot

  • New OpenRouter models: 4 (Inkling, Kimi K3, Muse Spark 1.1, Auto Router)
  • Limited free models: 5 (expiring July 21 through August 10)
  • Trending HuggingFace models: 10 new uploads
  • GitHub trending AI repos: 5 (767 to 350 stars)
  • AI news stories covered: 12 major stories (July 14-19)