Tech • AI • Robotics • Game
AI News: ChatGPT Images 2.6, Omni 1.2, Grok Bot Usage Slashed, Muse Voice and More!
8/10TL;DR
OpenAI, Google, Meta, Anthropic, xAI and Perplexity rolled out major AI updates this week, led by stronger image and video tools, new agent features and a broad shift in usage limits.
KEY POINTS
OpenAI adds GIF creation and reference images
ChatGPT Images 2.6 now supports animated GIF generation, with editable speed controls and follow-up image changes. Typical generations were reported at about 20 seconds, and the system performed well on character-style animations, though meme text rendering remained inconsistent.
OpenAI also added reusable image references, letting users save a person, pet or object under a name and call it back in later prompts without re-uploading. For developers, two new image models are coming to the API, with OpenAI claiming they are 50% faster.
Google upgrades its video model
Google released Gemini Omni 1.2 Flash, extending clip length to 15 seconds from 10 seconds. The company is also promising better consistency, lower cost and improved speed.
Demonstrations highlighted stronger identity preservation from reference images, more precise video editing, more natural-sounding voices and more cinematic camera control. The model is rolling out to Ultra and Pro users and is also available in Google Flow.
xAI faces backlash over tighter Grockbot limits
Users reported that Grockbot allowances were being exhausted far faster than before, with some saying usage jumped from roughly 50% to 100% unexpectedly. Elon Musk later said serving costs had become unsustainably high relative to revenue, indicating the tighter limits were intentional rather than a bug.
The change drew criticism because it arrived without advance warning, published replacement limits or any reset of depleted allowances. Despite the backlash, Grockbot is still seen by many users as one of the strongest AI agent platforms.
How users are cutting Grockbot consumption
The biggest savings appear to come from reducing automated routines, which consume allowance even when they do little or nothing. Examples included routines running 85 times per week or every 30 minutes, later cut back or disabled.
Other high-cost behaviors include bot group chats and delegated tasks between agents, as well as browser-based computer use. Directly using a single bot and favoring app connectors over cloud-browser actions appears to reduce burn significantly.
Meta expands Muse with voice and email-style features
Meta Muse is adding several capabilities, including a dedicated inbox-style feature often referred to as Musemail, plus broader voice interaction. Some users already have access to voice mode, which supports interruption, multiple voice options and an animated avatar during conversation.
Additional teased features point to a Muse Charm device and possible integration with Meta Glasses. The service is gaining attention partly because a $20 plan offers comparatively generous usage.
Anthropic loosens one Claude limit
Anthropic renamed Max 5X to Max and Max 20X to Ultra, while keeping prices at $100 and $200 per month. The more significant change is that Ultra accounts are no longer subject to the prior five-hour session limit.
Weekly limits still remain, but the removal of the shorter session cap marks a notable easing for heavy users. The change applies to the top-tier $200 plan.
OpenAI consolidates ChatGPT usage pools
OpenAI said ChatGPT and Codex will soon draw from the same allowance, ending a setup where basic chat usage remained separate while work and coding features already shared a pool. The company is also merging the chat and work tabs into a more automatic interface.
For heavy users, this reduces a common strategy of preserving advanced quotas by staying in regular chat whenever possible. OpenAI partially offset the shift by doubling the current work and Codex allowance for Pro users, but not for Plus subscribers.
Grockbot can now place phone calls
A new Grockbot feature lets the agent call businesses and individuals, identify itself as a virtual assistant and show a live transcript to the user. If the system lacks an answer mid-call, it can pause the interaction and ask the user for guidance before continuing.
Early tests showed fast responses and relatively natural speech, though with the polished tone of a call-center assistant. The feature extends AI agents from messaging and browsing into live voice task execution.
New avatar model raises realism concerns
A real-time avatar system called At Me V2 demonstrated near-live conversational performance with humanlike facial movement and speech. The company is pitching it for tutors, sales coaching, customer support and game characters.
Pricing claims drew particular attention: one comparison suggested the new model cost less than a cent, versus 50 cents for a comparable output from a rival system. Independent testing remains limited, and access is currently invite-only via API.
Finance tools and agent competition widen
Both ChatGPT and Claude expanded personal finance features in the United States, letting users connect accounts, review spending and detect duplicate or suspicious charges through read-only access. OpenAI also said ChatGPT Dots will gain more models, while Perplexity launched Teammates, background AI agents for Max subscribers.
The pattern across the market is clear: companies are converging on agent-style products that can monitor, call, summarize and act continuously in the background.
CONCLUSION
This week’s AI news combined eye-catching creative advances with a less welcome reality: tighter usage controls are becoming a central part of the market. As image, video and agent tools improve quickly, pricing and access policies may shape adoption as much as model quality.
Ask a question