Daily AI News | ChatGPT Free will default to Luna and add unlimited text chats
# Daily AI News | ChatGPT Free will default to Luna and add unlimited text chats
## Lead Story
### ChatGPT Free will default to Luna and add unlimited text chats
OpenAI said GPT-5.6 Luna will become the default model for ChatGPT Free and Go users this week, with unlimited text chats and a new Think button starting next week; Plus and Pro users can already access the updated GPT-5.6 Sol and a reasoning-effort slider. Unlimited text remains subject to abuse guardrails, while file uploads, images and other tools will keep their usage limits.
**AI take:** Core text access is becoming easier on the free tier, but unlimited text does not mean every feature is unlimited; files, images and tools still have quotas.
## More News
### Adobe launches one ChatGPT plugin spanning more than 70 tools
Adobe launched a unified ChatGPT plugin that brings more than 70 tools from Adobe Express, Photoshop, Firefly, Premiere, Acrobat, Lightroom, Illustrator, InDesign and Stock into one conversation, with global availability on ChatGPT web and apps. Users can access a basic tool set as guests, while signing in to an Adobe account unlocks generative features, Creative Cloud files and cross-session saving.
**AI take:** The unified plugin reduces app switching, but full generation, cloud-file access and saved work across sessions still depend on an Adobe sign-in.
### Ask Maps adds agentic food ordering, hotel and event discovery
Google added multi-step agentic capabilities to Ask Maps: it can find restaurants along a route and add a requested dish to the cart, compare live hotel prices and availability, and surface nearby events with ticket links; food ordering is rolling out first in the United States with partners including Square and Toast, with Uber Eats to follow. Users can also choose to connect Gmail so suggestions account for flights and restaurant reservations, with the connection off by default, while a new transit widget shows real-time delays.
**AI take:** Maps is shifting from a search surface toward an action surface; users still review and complete food orders, and Gmail personalization stays opt-in.
### Kimi K3 becomes generally available in GitHub Copilot
GitHub said Kimi K3 is generally available in Copilot and is rolling out gradually to Pro, Pro+, Max, Business and Enterprise plans across clients including VS Code, Copilot CLI, the cloud agent and mobile apps; Business and Enterprise administrators must enable it through policy first. Provider-list usage pricing is $3 per million input tokens, $15 per million output tokens and $0.30 per million cached input tokens; rollout was briefly paused during a GitHub Actions incident and has resumed.
**AI take:** Developers gain another agentic coding-model option, but enterprises do not get it by default; administrators still need to assess open-weight data-governance and compliance requirements.
### Google DeepMind open-sources its WeatherNext cyclone models
Google DeepMind released WeatherNext Cyclones. In evaluations on historical tropical cyclones from 2023 and 2024, the model gained more than 24 hours of lead-time accuracy on average across track, intensity and wind-structure forecasts: its three-day predictions matched the accuracy that previous models achieved at two days. The team also released code and weights for WeatherNext Cyclones, WeatherNext 2 and the smaller WeatherNext 2-mini, which can run on a single TPU, for research and operational evaluation.
**AI take:** An extra day of useful forecast time could materially improve preparation, but research results and open models do not replace official warnings from local weather agencies.
### Suno plans watermarks, audio fingerprints and tighter download rules
TechCrunch reported that Suno plans to use audio watermarking and fingerprinting on songs made with its platform to curb misuse through mass uploads to other streaming services. The company also plans a new download policy aimed at mass distribution and has partnered with Musixmatch to use its Sentinel copyright-detection system. Updated community rules explicitly prohibit deceptive audio presented as real and the unauthorized use of a real person's voice or likeness. Suno has not disclosed the watermarking system, rollout date or detailed download limits.
**AI take:** The plan moves provenance from principle toward product controls, but the launch timing, technical system and precise download boundaries remain undisclosed.
### AMD signs definitive deal to acquire Taalas for specialized AI inference silicon
AMD said it has signed a definitive agreement to acquire AI-chip startup Taalas. Taalas develops specialized technology that implements model-specific inference directly in silicon to improve performance and efficiency; AMD plans to bring its team and technology into the broader compute portfolio alongside the Instinct GPU roadmap. The announcement confirms the acquisition agreement and strategic direction, not a newly shipping product.
**AI take:** The deal shows AMD pursuing both general-purpose GPUs and model-specific inference silicon. The specialist approach may trade flexibility for efficiency, but customers do not yet have a new product or a stated availability date.
### DeepSeek signals a significant increase in API pricing
DeepSeek said in its official pricing documentation that overall API prices will rise in the near future and that the increase is expected to be significant. As of this report's cutoff, the company had not published the future adjusted rates, effective date or model-by-model increases; some reports linked the notice to V4's launch timing, but the notice itself did not confirm that link.
**AI take:** The direction of travel is clear, but the new rates and start date are not; developers should model budget scenarios instead of locking in costs from media speculation.
### Unitree prices its IPO at 150.8 yuan as DeepSeek joins the placement and model partnership
Exchange material shows Unitree priced its Shanghai IPO at 150.8 yuan per share, implying a valuation of about $9.04 billion and planned proceeds of 6.1 billion yuan. DeepSeek invested 140.8 million yuan in the strategic placement for 933,399 shares, or 2.31% of the placement allocation. The companies also agreed to jointly develop AI models for humanoid robots and give each other preference in model-training services, technical solutions, robot procurement and embodied-AI applications.
**AI take:** The IPO sets a public valuation benchmark, while DeepSeek's investment creates a route for model collaboration. Manufacturing scale, profitability and real-task performance still need separate proof.
### Lumilens raises more than $700 million in Series C funding at a $5.51 billion valuation
AI data-center optical-networking company Lumilens raised more than $700 million in Series C funding, taking total funding above $900 million at a $5.51 billion valuation, Reuters reported. The company is already shipping under hyperscaler contracts worth billions of dollars, confirming both financing and commercial delivery.
**AI take:** The combination of funding and shipments shows optical interconnect becoming a critical AI-infrastructure constraint and investment target. Production yield, expansion speed and customer concentration now matter more than valuation alone.
### Alphabet launches a bond sale of up to ten tranches, targeting $20 billion to $25 billion
Alphabet launched a dollar bond offering with as many as ten tranches, Reuters reported. People familiar with the matter put the fundraising target at $20 billion to $25 billion to support heavy AI investment. That range is a target rather than final proceeds and may change with demand and pricing.
**AI take:** A multi-tranche debt sale can spread the funding of large AI capital expenditure across longer maturities, but the target is not the final amount. Pricing, maturity mix and the eventual use of proceeds still matter.
## Rumors
### OpenAI and Jony Ive's first device is reportedly a portable, screenless smart speaker
The Verge and TechCrunch, citing Bloomberg's unnamed sources, reported that the first device being developed by OpenAI and Jony Ive's team could be a screenless, battery-powered smart speaker with a doughnut-like shape and roughly hockey-puck dimensions. It may include lights, cameras, other sensors and moving parts. The reports suggest a possible 2027 launch and a price of about $300 to $400; those details remain based on anonymous sourcing.
**AI take:** If accurate, this would be a premium portable, screenless AI device; its pricing and schedule could still change substantially.
### DeepSeek reportedly resumes talks for a nearly $8 billion round
Reuters, citing Bloomberg's unnamed sources, reported that DeepSeek may have resumed its second funding round, seeking close to $8 billion at a valuation of about 500 billion yuan, or roughly $74 billion. Monolith Management is reportedly discussing participation, but the round has not closed and the amount, valuation and investor list could still change.
**AI take:** If the round closes, DeepSeek could gain more room for compute and expansion; until agreements are signed, the target size, valuation and participants may all change.
