Tech • AI • Robotics • Game
AI News: $500 ChatGPT Plan? OpenAI Agent Leak & Opus 5.5
9.3/10TL;DR
Anthropic, OpenAI, and Meta unveiled lower-cost models, broader agent features, and new AI hardware, signaling a sharper industry push toward always-available personal assistants.
KEY POINTS
Anthropic cuts costs with Claude Opus 5.5
Claude Opus 5.5 launched as the first model in a new Claude 5.5 family, with 40% lower costs than Opus 5 while aiming for performance near Claude Fable 5.1. API pricing fell from $5 to $4 per million input tokens and from $25 to $20 for output tokens, with lower cached-input costs as well. Anthropic also said the model requires less compute to serve.
Benchmarks show stronger value, not just higher raw power
On reported benchmarks, Opus 5.5 outscored Fable 5.1 across coding, reasoning and workflow measures, including 66.4 on a coding benchmark versus 55.8. Gains in computer use were narrower, roughly 1 percentage point, underscoring that autonomous desktop-style task execution remains a harder frontier than coding. The model was also positioned competitively against GPT-6 Astra, trailing mainly in business workflows and agentic scientific research.
Speed and price matter more as enterprise use scales
Anthropic highlighted real-world efficiency tests rather than headline capabilities alone. In one code migration task, Opus 5.5 completed the work in 9.5 hours versus 12 hours for Fable 5.1, while costing 51% less. In another spreadsheet-and-presentation task, runtime dropped from 93 minutes to 63 minutes, with about 50% lower cost.
Anthropic targets readability as a product feature
A major focus of Opus 5.5 is cleaner communication. Outputs now prioritize important information first, reduce jargon, and follow user writing rules more closely. Practical examples showed summaries, bug explanations and design-change writeups formatted with clearer structure, bolded takeaways and shorter sections, an increasingly important distinction as users rely on AI for daily workplace decisions.
OpenAI expands GPT-6 lineup with cheaper models
OpenAI introduced GPT-6 Soul and GPT-6 Luna alongside GPT-6 Astra. Astra remains the premium option at $10 per million input tokens and $50 per million output tokens. Soul is priced at $2 input and $10 output for complex coding and professional tasks, while Luna is aimed at everyday scale usage at $0.10 input and $0.50 output, a sharp drop from top-tier pricing.
A $500 OpenAI plan is rumored
Code discovered by outside trackers pointed to a possible ChatGPT Pro Max plan priced at $500 per month. No official confirmation was available, but the leak suggests AI vendors may be testing a market for heavy users willing to pay well above the current premium tiers, especially for coding agents and higher usage limits.
Voice assistants are becoming app-connected agents
OpenAI also rolled out a new voice mode powered by GPT-6 Astra, Soul, and Luna that can use plugins for services such as email, calendar, and Slack across web, mobile, and work products. The system responds quickly in conversation, though pulling information from connected apps can still take 15 to 20 seconds, showing progress but also latency limits in live agent workflows.
Meta enters the hardware race with Muse Charm
Meta announced Muse Charm, a small physical AI device designed to run the MetaMuse assistant. The product is intended for tasks such as sending emails, organizing a day, and planning travel through voice interaction. Mark Zuckerberg said only a small number of prototypes currently exist, with a December shipping target and no announced price.
The wider race is shifting toward personal AI assistants
AI companies are increasingly building cross-app agents rather than standalone chat tools. A leaked Tesla Cybertruck demo showed Grokbot handling inbox, calendar and file-management tasks, while one company said its support operation handled a 175% increase in tickets without adding staff and may have otherwise needed 200 more hires.
OpenAI agent rumors point to a social layer
Industry speculation centers on a possible new OpenAI agent platform, reportedly codenamed Aeon and possibly linked to a product name such as Orbit. The rumored system could include its own profile, direct messages and group features, placing it in more direct competition with Grokbot and MetaMuse. If launched soon, it would mark a step toward AI agents that are not only task tools but persistent digital counterparts.
CONCLUSION
The week’s announcements show the AI market moving beyond pure model intelligence toward cheaper inference, stronger usability, and always-on agents tied to apps and hardware. The next competitive battleground is increasingly clear: whoever delivers the most useful personal assistant at sustainable cost may shape the next phase of consumer and workplace AI.
Ask a question