Gemini price war, Shazeer jumps to OpenAI, and AI agents invade Creative Cloud

Kimi K2.7 Code: Moonshot AI’s open-source challenger to GPT-5.5

Moonshot AI has released Kimi K2.7 Code, a one-trillion-parameter open-source mixture-of-experts model designed for complex coding tasks with multimodal capabilities and agentic programming. It excels in dynamic web component generation and front-end design, processing text, code, and visual elements. The model integrates multi-step tool usage to autonomously execute workflows and shows strong results in long-context tasks and independent evaluations like the Aeros Smoke Test. However, its context window is limited to 262,000 tokens, and token efficiency is 260 tokens per second in high-speed mode, which raises costs in high-volume scenarios. Pricing is set at $0.19 per million input tokens for cache hits and $0.95 for cache misses, with output tokens at $4 per million. A quantized version reduces the model size to 325 GB for easier deployment, and integration with Docker Sandbox enables secure execution. Compared to proprietary models like Opus 4.8 and GPT-5.5, Kimi K2.7 Code is more cost-effective but still lags in refinement and overall performance. Its predecessor was Kimi K2.6, and an upcoming Kimi K3.0 is expected to address token efficiency, context window size, and performance.

Kimi K2.7 vs ChatGPT 5.5 : What Makes Kimi K2.7 Code a Threat to GPT 5.5 →

Adobe rolls out AI agents across Creative Cloud apps

Adobe is launching its “creative agent” across Creative Cloud and third-party AI platforms including ChatGPT and Claude. The agent appears as an AI Assistant in Premiere, Photoshop, Illustrator, InDesign, and Frame.io in public beta, with After Effects in private beta. Each assistant is tuned to its specific app, handling repetitive production tasks like sorting footage, batch-renaming clips, swapping backgrounds, or generating multiple versioned files from a spreadsheet. In Frame.io, it organizes footage and pulls together feedback. The agent is designed to handle multi-step workflows while users retain creative control. Separately, Adobe is testing a redesigned Firefly Studio interface in private beta that combines generation and editing, along with new features like brand kit tools, product photo-to-video, Quick Cut, and Elements for reusable assets. Adobe’s tools already integrate with OpenAI’s ChatGPT, Anthropic’s Claude, and Microsoft 365 Copilot, with Google Gemini and Slack integrations coming. Forest Key, who runs agentic AI at Adobe, noted that creative ideas rarely start in one app and users should not have to switch tools to act on them.

Adobe adds AI agents to Photoshop, Premiere, and more Creative Cloud apps →

Google slashes Gemini subscription to $4.99 per month

Google reduced the price of Google AI Plus from $7.99 to $4.99 per month on June 18, 2026. The subscription now includes 400 GB of cloud storage across Gmail, Drive, and Photos, doubling the previous amount. The plan offers access to Gemini 3.1 Pro, Deep Research, Gemini in Google apps, and image, music, and video generation models, plus credits for Google Flow. Family sharing with up to five others is supported. The price cut directly targets OpenAI’s ChatGPT Go tier, which costs $8 per month and provides access to GPT-5.5 Instant and other features.

Google just went nuclear on OpenAI with a surprise $4.99 Gemini price cut →

Noam Shazeer, Gemini co-lead, leaves Google for OpenAI

Noam Shazeer, co-author of the influential AI paper “Attention Is All You Need” and Vice President of Engineering at Google where he co-led the Gemini models, is leaving Google to join OpenAI. Shazeer originally joined Google in 2000 and left in 2021 to co-found Character.AI. He returned to Google in 2024 as part of a $2.7 billion deal to improve Google’s reasoning models. “It was a difficult decision to move on,” Shazeer wrote. His move is described as the biggest AI talent story of the year, alongside Andrej Karpathy joining Anthropic. The article is dated June 18, 2026.

Google’s Gemini co-lead Noam Shazeer joins OpenAI after two-year return stint →

Amazon, Nvidia, and AMD invest $310M in world model startup Odyssey

Amazon, Nvidia, and AMD each contributed through their venture arms to a $310 million investment round in AI startup Odyssey ML, valuing the company at $1.45 billion. Odyssey builds world models that simulate the physical world in 3D, including physics, body language, and dynamics. Other backers include the CIA-linked fund IQT, GV, Google chief scientist Jeff Dean, and investor Elad Gil. Odyssey uses Amazon Web Services as its preferred cloud provider and runs on Amazon’s Trainium chips. The company employs 55 people across London, Zurich, and Palo Alto. World models are seen as the next major bet after language models, with Meta’s Yann LeCun arguing they are necessary for human-level intelligence and Google DeepMind’s Demis Hassabis calling them key to general AI. AI pioneer Fei-Fei Li is pursuing the same concept at her startup World Labs.

Amazon, Nvidia, and AMD bet $310 million on AI startup building 3D world models →