Two AI giants, OpenAI and Anthropic, have recently come under fire for allegedly implementing covert performance reductions on their flagship models, triggering strong backlash from users. It has been revealed that OpenAI is grayscale testing a new version called GPT-5.6-sol on the Codex platform, significantly reducing inference compute by lowering the Juice value from 768 to 128 — an 83% cut. At the same time, a large number of users have accused Anthropic of silently downgrading Claude Opus 4.8, causing noticeable degradation in core capabilities such as logical reasoning and contextual memory.
Background: Covert Performance Cuts by AI Leaders
OpenAI and Anthropic, both leaders in generative AI, are facing severe capital pressures. Reports indicate that OpenAI is testing the GPT-5.6-sol model version on Codex, which slashes the inference compute (Juice value) from the original 768 to 128 to reduce the cost per inference call. While this strategy helps OpenAI control operational expenses while maintaining API services, it directly degrades output quality, particularly in complex reasoning tasks.
OpenAI’s GPT-5.6-sol Test: 83% Compute Reduction
Specific data shows that reducing the Juice value from 768 to 128 represents an ~83.3% cut in inference compute. Codex users have observed that response times of GPT-5.6-sol shortened by approximately 40% under identical inputs, but the frequency of logical leaps and errors in answers increased significantly. OpenAI has not yet issued an official statement regarding this grayscale test, but market analysts believe the move is likely related to delays in the company’s IPO process and increased scrutiny from capital markets on AI companies’ profitability.
Anthropic Accused of Silent Downgrade on Claude Opus 4.8
Meanwhile, Anthropic is not immune. A surge of user complaints on social media and technology forums indicates that Claude Opus 4.8 has experienced an unexplained “dumbing down”: its performance in long-text reasoning, multi-step logic tasks, and contextual memory has notably deteriorated. Users comparing historical conversation logs have found that responses from early April were far superior to recent ones. Although Anthropic has not officially confirmed the downgrade, users widely suspect it is a measure to cut operational costs. This “silent downgrade” has eroded trust.
Underlying Causes: IPO Obstacles and Capital Winter
Analysts point out that both OpenAI and Anthropic share a common dilemma: high compute investment versus limited market monetization. Both companies have reportedly planned IPOs but face slow progress due to market conditions and regulatory pressures. In the ongoing capital winter, venture capital firms are more cautious about AI startup valuations, forcing companies to reduce costs to extend runway. Directly lowering model quality has become the most covert and expedient method, albeit at the cost of user trust.
User Trust Crisis and Industry Impact
The outcry for “refunds” reflects widespread user dissatisfaction. This incident has far-reaching implications for the AI industry. First, users are demanding greater transparency in model version updates, calling for established performance benchmarks and change announcement mechanisms. Second, regulators may begin investigating whether such silent downgrades violate consumer protection laws. Finally, smaller AI firms may seize the opportunity to highlight their own model stability and service quality to capture market share. As of now, neither OpenAI nor Anthropic has issued a formal response, but the debate continues to escalate within the industry.