It is still far from truly replacing programmers: AI agents accessing enterprise private code only achieve a maximum pass rate of 40%.
Insight Beating AI News Flash: Y Combinator (YC)-incubated AI data firm Specific Labs has launched Real-SWE, a coding evaluation benchmark that tests coding agents on real private code from actual enterprises. The tasks are sourced directly from production environments, including tax calculation, bill migration, API billing, and customer data migration. Neither the code nor answers are publicly available, so agents must independently master the company’s business rules and code structure. Results show Fable 5.1 ranked first with a 38.8% pass rate; GPT-6 Astra followed at 33.8%, Gemini 3.8 Flash at 31.2%. GLM 5.3 took fourth place with 28.8%, outperforming Grok 4.6, Kimi K3, and GPT-5.6 Sol. Of the 10 currently public tasks, 6 have an overall pass rate below 15%, and one task saw all models fail. The standout performer was GLM 5.3. Real-SWE assesses agents’ ability to continuously read code, identify rules, and complete multi-step modifications in unfamiliar projects. Zhipu AI researcher Xiaopu Peng noted that his team has been training long-range tasks since GLM-5.1, and Real-SWE is more aligned with such capabilities than public SWE benchmarks, which partly explains GLM 5.3’s strong performance here.
5 minutes ago
Ahead of the CPI release, Tom Lee reiterates his bullish September view: Macroeconomic data will completely ease inflation concerns, pushing the probability of a September interest rate hike to zero.
Tonight, the market will get the last key macroeconomic indicator ahead of the Federal Reserve’s interest rate decision: the CPI data. Markets view this as the most critical signal for the Fed’s policy announcement. Prior to that, Bitmine Chairman Tom Lee has reaffirmed his bullish outlook for September. In early September, Lee told CNBC that he was optimistic about U.S. stock market performance in the month. He argued that as employment data, CPI, and the Fed’s interest rate decision are released sequentially, market concerns over inflation may ease. While investors remain cautious now, this has formed a classic "wall of worry"—a historical pattern that often marks a buying opportunity.
5 minutes ago
Hyperliquid’s protocol revenue over the past 30 days is approximately $59.84 million, with 775,400 HYPE tokens burned in the same period.
According to TradingBeats' monitoring, Hyperliquid's protocol revenue over the past 30 days reached approximately $59.84 million, an 83.3% increase from the prior 30-day period. However, revenue over the most recent seven days dropped to roughly $11.57 million, a 12.7% week-over-week decline, marking notable expansion in monthly revenue though a recent cooling in activity. Meanwhile, between August 12 and September 11, the Assistance Fund’s cumulative burned HYPE tokens rose to 47.123 million, with around 775,400 new HYPE burned over the past 30 days, valued at approximately $62.08 million. Per the official mechanism, the Assistance Fund automatically converts trading fees allocated to it into HYPE, and the HYPE held in the fund is permanently removed from circulation and the total token supply.
5 minutes ago
Gulf Transport Disruptions Persist, Wall Street Raises Oil Price Forecasts
According to market reports, major Wall Street banks are raising their Brent crude oil price forecasts amid ongoing Middle East shipping disruptions that have exacerbated global oil supply tightness. HSBC projects Brent crude’s 2026 average price at $90, while Goldman Sachs warns prices could exceed $120 if severe production disruptions persist in the Gulf region. Bank of America forecasts an even more extreme scenario, with prices potentially hitting $150 if energy infrastructure suffers major damage.
5 minutes ago
OpenAI Denies Misappropriating Mathematician’s Unpublished Research: Codex Draft Did Not Impact Navier-Stokes Proof
Beating AI Express News: OpenAI has responded to the Navier-Stokes credit dispute, explicitly denying it improperly used unpublished research from New York University mathematician Tristan Buckmaster. Following an internal investigation, the company stated that prompts Buckmaster submitted to Codex over the past two months could not have impacted its internal model in any way, including via training. Buckmaster and Levent Alp?ge had been working on the same research area, uploading numerous drafts to Codex. After their progress reached OpenAI, the company deployed concentrated computing power to produce a Navier-Stokes proof within days. Buckmaster thus suspected his unpublished research had been incorporated into the model, a claim OpenAI has now explicitly rejected. OpenAI also set a timeline boundary at July 3, noting that any user input after that date could not affect the system. Some community members have speculated this date may correspond to a training data cutoff or model checkpoint, though OpenAI has not provided an explanation. The key breakthrough by Buckmaster and Alp?ge occurred in mid-August, far later than the July 3 boundary set by OpenAI.
5 minutes ago
App Store
Profile
Security
Sign Out
Feeds
Articles
Source
Add to Favorites
Download image
Share x
Copy link
