Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingOpenAI fixes three major flaws in the GPT-6 Astra model and resets the usage limits for Codex subscribers.
Core points:
- Astra users have released a series of comparison tests in the past week, claiming that the model has been quietly weakened since its release.
- OpenAI's Codex leader pointed to three causes of the problem, including a context management experiment that affected approximately 4,000 to 5,000 users.
- The company's previous flagship product, GPT-5.6 Sol, faced similar charges in July.
OpenAI fixes quality flaws in Astra
Tibo Sottiaux, Codex product leader, said the team worked with users to trace the root cause of the failure, then released a fix and fully reset usage restrictions before midnight local time. According to updates released on Saturday, the main problems include the following:
First, skills written for early models are triggered too frequently, sometimes preventing Astra from checking its own work. Second, an optional context management experiment could cause the model to prematurely terminate or reply to outdated information; OpenAI has disabled the experiment after affecting approximately 4,000 to 5,000 users. In addition, some engines are misconfigured, resulting in a measurable degradation in the quality of tail traffic routed through them. To this end, the company removed the engines and made a number of minor fixes. Sottiaux pointed out that the current model can more accurately track users 'latest news.
Astra users report retrogression in coding capabilities
Complaints continue to ferment on social media platform X. The developers ran the same prompt words on the release day version of Astra and the current version, leaving all settings unchanged, and compared the results. Pranjal Paliwal, the developer who praised the model a few days ago, reviewed the code the model wrote for him and characterized the result as a "regression" rather than an improvement.
Dax Raad, builder of the coding tool Opencode, said he had switched the team back to the older GPT-5.6 Sol model because the cost of using Astra doubled and the negative impact was not considered worth it. A forum poster described that Astra's performance was on par with Sol's under the same task and the number of retries was the same. However, not everyone accepts this view. One user using the pseudonym Antikythera argued that the model has always been lazy and likes to use bullets, so users are just no longer surprised by it.
Calculating limits loom over the prospects of GPT-6 Astra
Capacity issues have plagued its promotion process since the model became widely available in early September. Users reported that OpenAI has cut Astra usage limits for heavy ChatGPT users and suspended applications for new $200 Pro subscriptions, a move confirmed by Sottiaux on September 10. Currently, the model still costs $10 per million of input tokens and $50 per million of output tokens, which is 2.5 times the price when Sol was released.
This is the second OpenAI flagship product in two months that has been mired in controversy after paying users filed the same allegations. In July, when users reported that Sol's top reasoning model had become shallow overnight, Sottiaux denied the idea of deliberate weakening, while confirming that the company had been conducting experiments on reasoning efforts.
