WireTensors

Updated Sun, 01 Nov 2026 23:39:47 UTC

OpenAI Doubles ChatGPT Pro Limits as GPT-6.1 Terra Launches; Anthropic Patches Watermark Bypass

Roundup facts
Published 2026-11-01
Items 6
Coverage Writing, coding, image, video, productivity, SEO
Last verified 2026-11-01

OpenAI's About-Face: ChatGPT Pro Limits Double Instead of Halve

OpenAI scrapped a planned 50% reduction in usage caps for ChatGPT Pro subscribers and instead doubled limits across the $100, $200, and $500/month tiers, plus Business Premium. This reversal signals OpenAI's confidence in scaling capacity—or sensitivity to subscriber backlash over constrained access. For power users, the move materially changes the value proposition of premium tiers; for OpenAI, it's a bet that higher volume justifies the infrastructure cost. The timing coincides with intensifying competition from Claude and other models, making retention critical.

GPT-6.1 Terra Targets Visual Reasoning and Computer Use at New Price Point

OpenAI's latest model release, GPT-6.1 Terra, focuses on visual reasoning, computer use, and technical communication, priced at $1 per million input tokens and $5 per million output tokens—a significant undercut versus prior flagship models. Early reports highlight improved computer-use performance, positioning Terra as a direct competitor to Anthropic's Claude Code and Microsoft's desktop agents. The emphasis on vision and autonomous task execution mirrors industry momentum toward agentic systems; enterprises evaluating multi-modal and reasoning-heavy workflows should factor Terra into cost-benefit analysis against Claude and open alternatives.

Microsoft's Mage-VL-Cowork Cuts Model Calls by 13% in Desktop Automation

Microsoft released Mage-VL-Cowork, a 4-billion-parameter vision-language model, which scored 66.0% on OSWorld-Verified whilst reducing model calls by 13.0% and output tokens by 25.4% compared to earlier approaches. The efficiency gains matter for cost-sensitive deployments of autonomous desktop agents—shrinking both latency and token burn. This development underscores a broader trend: as agent systems mature, the engineering focus shifts from raw capability to inference efficiency and multi-step coordination, favoring leaner models over brute-force scaling.

Anthropic Patches Watermarking Flaw as EU AI Act Deadline Looms

Anthropic extended its text-watermarking system to four additional Claude models ahead of the EU AI Act's December 2026 compliance deadline. However, researchers have identified a sampling-parameter workaround that reportedly bypasses watermarking on Opus 4.6 and Sonnet 4.6. The discovery highlights the fragility of watermarking as a regulatory safeguard and raises questions about whether Anthropic's fix will hold or invite further attacks. For enterprise users in the EU, the vulnerability and its fix exemplify the real-time cat-and-mouse dynamics between compliance obligations and adversarial researchers.

DeepSeek Enters Autonomous Coding-Agent Space

DeepSeek, previously focused on model development, released a prototype autonomous coding agent in direct competition with Anthropic's Claude Code. The move mirrors a broader shift—model providers are bundling end-user applications alongside APIs to capture value across the stack. DeepSeek's entry signals that coding agents are now table-stakes for major model labs, intensifying competition on both quality and user experience.

Mistral Releases 1-Trillion-Parameter Open Weights (With Caveats)

Mistral published open weights for Mistral Large 4 on 29 October, marketed as a 1-trillion-parameter multimodal model but with active-parameter counts reported between 49 billion and 52 billion—likely via mixture-of-experts sparsity. Training reportedly consumed 4,000 Nvidia Grace Blackwell GPUs. For open-source practitioners, the release offers a high-capacity alternative to closed models, though discrepancy between parameter marketing and actual active compute raises questions about true capability parity and inference cost claims.

Roundup FAQ

What is this roundup? +

OpenAI reversed course on a planned usage cut, doubling ChatGPT Pro caps across all tiers whilst launching GPT-6.1 Terra—a vision-focused model priced at $1–$5 per million tokens. Meanwhile, Anthropic scrambles to fix a watermarking vulnerability ahead of the EU AI Act's December deadline, and Mistral releases open weights for a 1-trillion-parameter multimodal model.

When was it published? +

This roundup was published and verified on 2026-11-01.

What topics does it cover? +

It covers: OpenAI's About-Face: ChatGPT Pro Limits Double Instead of Halve; GPT-6.1 Terra Targets Visual Reasoning and Computer Use at New Price Point; Microsoft's Mage-VL-Cowork Cuts Model Calls by 13% in Desktop Automation; Anthropic Patches Watermarking Flaw as EU AI Act Deadline Looms; DeepSeek Enters Autonomous Coding-Agent Space; Mistral Releases 1-Trillion-Parameter Open Weights (With Caveats).

Is the coverage neutral? +

Yes. Roundups summarise developments neutrally and do not promote any single vendor.

Does this roundup contain affiliate links? +

Links within roundups may be affiliate links; we may earn a commission at no extra cost to you, and this never affects coverage.

Where does the information come from? +

Roundups summarise vendor product pages, changelogs and public announcements, each verified on the publication date.

How often are roundups published? +

WireTensors aims to publish short roundups on a daily cadence.

Where can I read full tool reviews? +

Each tool mentioned has a full review under /tools, with pricing, ratings, pros, cons and FAQs.

Reviewed by Arjun Mehta

Editorial lead overseeing WireTensors' research, sourcing and verification process

Last verified:

Sources