AI News Feed
Market watch
Xbox Elite Controller 3 Specs Reportedly Leak From Xbox Design Lab BackendAsos Investigates App Notification Claiming Hack as Shares Fall Almost 12%Amazon October Prime Day discounts smart home gear, headphones, docks and speakersscrcpy 5.0 Adds GPU Video Decoding and Native Windows on ARM64 BuildsCornell Ecologist Aims to Define 'Smoke Ecology' as Wildfire Smoke Affects BirdsHack The Box launches AI Range Enterprise Edition to test whether AI agents can do the job
Xbox Elite Controller 3 Specs Reportedly Leak From Xbox Design Lab BackendAsos Investigates App Notification Claiming Hack as Shares Fall Almost 12%Amazon October Prime Day discounts smart home gear, headphones, docks and speakersscrcpy 5.0 Adds GPU Video Decoding and Native Windows on ARM64 BuildsCornell Ecologist Aims to Define 'Smoke Ecology' as Wildfire Smoke Affects BirdsHack The Box launches AI Range Enterprise Edition to test whether AI agents can do the job
Companies

OpenAI's 28-Day Sprint Starts With Speed Gain, but Ads, Watermarks and Subscription Value Stir Backlash

OpenAI's 28-day sprint opened with about 50% faster GPT-6 Astra and Sol, but ads, EU watermarks and subscription cuts drew backlash.

AI News Feed · 2026-10-06 18:11

Tibo, the head of Codex, said on X that over 28 days OpenAI would either deliver one Codex or Work improvement each day or issue a full quota reset. A full reset was framed as a penalty avoided only by shipping improvements. The first delivery was speed optimization. The company said the faster inference applies to official subscriptions and third-party ecosystems including OpenCode, Pi, Amp and Devin.

The focus on speed followed a troubled launch. GPT-6.1 Sol, released on Sept. 29, was widely slow even in Fast mode, and users joked it should be renamed GPT-6.1 Turtle. Tibo apologized, saying the model had faced a huge load increase in its first two days, and announced restored expected speeds and quota resets for paid accounts. He then sought user feedback and set four priorities: simplifying products, improving efficiency to support more usage, breakthrough features and new models. The 28-day plan followed.

At the same time, OpenAI announced two changes that alarmed workers and developers. It said ChatGPT, which reaches 1.2 billion users weekly, would test a new visual ad format during image generation. Sponsored brand images would appear while users wait for image output, with a "Learn more" button. OpenAI said it is working with dozens of attribution and measurement platforms to help advertisers calculate ROI and incremental conversions, according to QbitAI.

OpenAI also said that within weeks it would add invisible statistical watermarks to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act. The company acknowledged significant limitations in current text watermarking technology but said it would embed the invisible statistical signal in generated text streams for compliance.

A report from independent research and analysis firm SemiAnalysis then intensified criticism of OpenAI's subscription economics. The report said monthly subscriptions are heavily subsidized loss leaders. At Anthropic, subscription fees accounted for only 10% of revenue but consumed more than 40% of inference compute, reducing revenue per megawatt by about $36 million. The two companies took different approaches under compute cost pressure. Anthropic quietly reduced the API-equivalent value of high-end models such as Fable 5.1, while OpenAI cut the API-equivalent value of its $200 Pro plan by half and introduced a $500 subscription tier.

According to SemiAnalysis, the new $500 plan offers only 21% more GPT-6 Astra quota than the old $200 plan. Because OpenAI also cut the cached input price for 6.1 Sol, the API-equivalent value of Sol-level models did not increase and even declined under the new plan. At Anthropic, the $200 tier's Opus 5.5 was estimated to provide about five times the API-equivalent value of OpenAI's corresponding model, and Anthropic also led on mid-tier models commonly used by subscribers. The report said OpenAI had previously attracted independent developers and engineers from Claude with relatively generous subscription quotas, but the latest changes erased that advantage.

User reactions to the first day were mostly negative. A few users said the 28-day challenge had started well. Others tested the speed gain and disputed OpenAI's claims. One user, @stalkermustang, said that after four hours speed barely reached 35 TPS, below Tibo's claim of reaching 50 TPS within two hours. Another, @V3zxS8myOU4056, wrote: "Show benchmarks. Do you think we are idiots?" Some users said the acceleration was compensating for the slowdown after GPT-6.1 Sol launched and had not returned to pre-launch speed. Others said GPT-6.1 Sol remained slower than Gemini 3.8 Flash and Opus 5.5. One user asked whether faster model speed would also drain quotas faster.

The first day of the 28-day marathon thus ended with the announced speed improvement overshadowed by complaints about advertising, watermarks and subscription value. QbitAI reported that users questioned whether OpenAI, once known for its nonprofit roots, was now cutting developer quotas, pushing a $500 plan and inserting ads. The report said many users were not buying the combination and called on Tibo to focus on core model performance and restore full quota resets.

Editor's Summary

OpenAI's 28-day improvement sprint began with a roughly 50% speed increase for GPT-6 Astra and GPT-6.1 Sol, but the first day drew backlash over advertising tests in ChatGPT, EU text watermarks and subscription value. A SemiAnalysis report said OpenAI cut the API-equivalent value of its $200 Pro plan and introduced a $500 tier that offers only 21% more GPT-6 Astra quota, while Anthropic's $200 Opus 5.5 offers about five times the API-equivalent value. Users disputed the claimed speed gains and questioned whether faster inference would consume quotas more quickly.