DeepSeek unveils upgraded giant-challenging R1 model
Shafaq News/ Chinese AI firm DeepSeek has released an upgraded version of its flagship R1 reasoning model, intensifying competition with US leaders OpenAI and Google.
The updated model, R1-0528, significantly enhances performance in complex inference tasks, narrowing the gap with OpenAI's o3 series and Google's Gemini 2.5 Pro, according to a post on the developer platform Hugging Face.
While described as a 'minor' version upgrade, R1-0528 introduces substantial improvements in mathematical reasoning, programming, and logical deduction. DeepSeek also reported a 50% reduction in hallucinations—AI-generated false or misleading output—in tasks such as rewriting and summarization.
In a WeChat post, the Hangzhou-based firm said the model now excels at generating front-end code, roleplaying, and producing creative writing including essays and novels. 'The model has demonstrated outstanding performance across various benchmark evaluations.'
Originally launched in January, R1 quickly went viral, challenging assumptions that advanced AI development requires vast computing infrastructure. Its success triggered responses from Chinese tech giants such as Alibaba and Tencent, both of which released competing models claiming superior performance.
DeepSeek also disclosed that it applied a distillation technique—transferring the reasoning methodology from R1-0528—to enhance Alibaba's Qwen 3 8B Base model, boosting its performance by more than 10%.
'We believe the chain-of-thought from DeepSeek-R1-0528 will hold significant importance for both academic research and industrial development focused on small-scale models,' the company added.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Shafaq News
13 hours ago
- Shafaq News
Iraqi MP accused of exploiting oil job Issue for votes
Shafaq News/ On Saturday, an Iraqi labor rights official accused members of parliament in Basra of exploiting a long-standing employment issue in the oil sector to boost their campaigns ahead of Iraq's upcoming elections. Speaking to Shafaq News, Hakim al-Aidani, director of the Center for the Defense of Workers in Oil Fields, pointed to the renewed spotlight on the long-demanded 'safety passport,' a certification required by many seeking jobs in the province's oil industry. 'The issue first emerged when BP began operating in the fields,' al-Aidani recalled. 'Thousands of young people appealed for action, but the file remained ignored for years.' The current version of the document, known locally as the 'red passport,' is not accepted by oil companies, even for entry-level jobs such as cleaners at fields like Rumaila. Job seekers who obtain it are often rejected when they apply. Al-Aidani noted that Total, which now oversees multiple projects in Basra, requires a different type of certification. This shift follows earlier patterns, including a period when a Chinese operator invalidated the previously used 'green passport.' Each transition, he added, renders older documents useless and forces applicants to restart the process. He also questioned the move to offer the passport free of charge, after years during which applicants paid more than $500 to obtain it. 'Where were officials when young people were bearing these costs?' he asked.
Shafaq News
a day ago
- Shafaq News
Oil finishes down on possible OPEC+ output hike
Shafaq News/ US crude futures fell on Friday as traders expected OPEC+ would decide on Saturday to boost oil output for July beyond previous forecasts. Brent crude futures settled down 25 cents, or 0.39%, at $63.90 a barrel. US West Texas Intermediate crude finished down 15 cents, or 0.25%, at $60.79 a barrel, having earlier dropped more than $1 a barrel. The Brent July futures contract is due to expire on Friday. The more liquid August contract was down 71 cents, or 1.12%, at $62.64 a barrel. At these levels, the front-month benchmark contracts were headed for weekly losses over 1%. Prices dipped into negative territory after Reuters reported that OPEC+may discussan increase in July output larger than the 411,000 barrels per day (bpd) rise that the group decided on for May and June. "What OPEC+ is planning doesn't look particularly supportive for the oil market," said Matt Smith, Kpler's lead analyst for the Americas. The potential OPEC+ output hike comes as the global surplus has widened to 2.2 million bpd, likely necessitating a price adjustment to prompt a supply-side response and restore balance, said JPMorgan analysts in a note, adding that they expected prices to remain within the current range before easing into the high $50s by year-end. Phil Flynn, a senior analyst with Price Futures Group, said an online post on Truth Social by US PresidentDonald Trumpthat seemed to threaten more changes in tariff levels for Chinese imports also put pressure on crude prices. "Trump's Truth Social message on China failing to observe a truce on tariffs also combined with the Reuters headline to push prices down," Flynn said. Trump's tariffs were expected to remain in effect after a federal appeals court temporarily reinstated them on Thursday, reversing a trade court's decision a day earlier to put animmediate blockon the sweeping duties. US energy firms this week cut the number of oil and natural gas rigs operating for a fifth week in a row to the lowest since November 2021, energy services firm Baker Hughes(BKR.O), opens new tabsaid in its closely followed report on Friday. It was the first time since September 2023 that the number of rigs declined for five straight weeks. Baker Hughes said this week's decline put the total count down by 37 rigs, or 6%, from this time last year. Oil rigs fell by four to 461 this week, their lowest since November 2021, the company said. Gas rigs rose by one to 99.
Shafaq News
2 days ago
- Shafaq News
DeepSeek unveils upgraded giant-challenging R1 model
Shafaq News/ Chinese AI firm DeepSeek has released an upgraded version of its flagship R1 reasoning model, intensifying competition with US leaders OpenAI and Google. The updated model, R1-0528, significantly enhances performance in complex inference tasks, narrowing the gap with OpenAI's o3 series and Google's Gemini 2.5 Pro, according to a post on the developer platform Hugging Face. While described as a 'minor' version upgrade, R1-0528 introduces substantial improvements in mathematical reasoning, programming, and logical deduction. DeepSeek also reported a 50% reduction in hallucinations—AI-generated false or misleading output—in tasks such as rewriting and summarization. In a WeChat post, the Hangzhou-based firm said the model now excels at generating front-end code, roleplaying, and producing creative writing including essays and novels. 'The model has demonstrated outstanding performance across various benchmark evaluations.' Originally launched in January, R1 quickly went viral, challenging assumptions that advanced AI development requires vast computing infrastructure. Its success triggered responses from Chinese tech giants such as Alibaba and Tencent, both of which released competing models claiming superior performance. DeepSeek also disclosed that it applied a distillation technique—transferring the reasoning methodology from R1-0528—to enhance Alibaba's Qwen 3 8B Base model, boosting its performance by more than 10%. 'We believe the chain-of-thought from DeepSeek-R1-0528 will hold significant importance for both academic research and industrial development focused on small-scale models,' the company added.
