|
|
|
Aug 03, 2026
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
Welcome back! OpenAI previews a new series of models, tentatively named "Astra," in Washington D.C. DeepSeek's smaller V4-Flash model makes a splash. Amazon completes the remaining $35 billion investments in OpenAI.
|
|
|
|
OpenAI is preparing to release a new model family, tentatively using the name “Astra,” with improved abilities to complete long-running tasks, according to three people briefed on the plans. OpenAI CEO Sam Altman demonstrated Astra to policymakers and regulators in Washington, D.C., this week, according to the people. OpenAI touted its abilities to have multiple agents work together over a long period of time to solve particularly hard problems, which could be applied to a project or advanced math problem. Astra would be a new class of OpenAI models, alongside Sol, Terra and Luna. It couldn’t be learned how soon OpenAI plans to release it, according to the people. OpenAI also hasn’t decided whether Astra would be labeled GPT 6, or an additional model in the GPT 5 series, such as GPT 5.7. While the models are already in testing, they are intended to be the first to go through the Trump administration’s planned new framework for submitting AI models to the federal government before releasing them to the public, according to one of the people. The administration is aiming to finalize that framework by a self-imposed deadline this weekend. In the short term, OpenAI also plans to release a report about how OpenAI was able to solve 10 previously unsolved math problems, demonstrating the capabilities of its most advanced AI, the people said.
|
|
|
|
Chinese AI developer DeepSeek’s release on Friday of a smaller version of its flagship V4 open-source model is causing a stir. In early tests, the model, V4-Flash, demonstrated capabilities comparable to much larger open-source and proprietary models, and it’s also less expensive than those rivals. People can buy access to it through DeepSeek’s cloud-based application programming interface or download the model and use it on their own servers. The new V4-Flash model isn’t as powerful as Moonshot AI’s Kimi K3 on benchmarks, but could tap into a growing market for affordable, smaller models that can power autonomous agents without breaking the bank. V4-Flash appears to rival OpenAI’s GPT-5.6 Luna and Chinese firm Zhipu AI’s popular GLM-5.2 open-source model, according to Artificial Analysis Intelligence Index, a comprehensive benchmark for AI models’ capabilities across math, science, coding and reasoning. V4-Flash’s cost per task can be 60% lower than that of GPT-5.6 Luna, according to Artificial Analysis. DeepSeek is expected to release the official version of V4-Pro, a larger version of V4 with six times as many parameters, or settings in the model, soon.
|
|
|
|
Amazon has completed its $50 billion investment in OpenAI, putting in the remaining $35 billion in two stages in recent months, the commerce-and-cloud firm revealed in a securities filing on Friday. Amazon agreed in late February to invest $50 billion into the ChatGPT creator, but it only put in $15 billion immediately, with the rest to come either when OpenAI goes public or when OpenAI meets “specified milestones.” Friday’s securities filing said that Amazon put $13.7 billion of the $35 billion into OpenAI in the second quarter and the remaining $21.3 billion since the end of June. The filing didn’t say what prompted Amazon to make the remaining investment, as OpenAI hasn’t gone public.
|
|
|
|
A new venture capital fund called SemiAnalysis Capital Fund I, managed by Dylan Patel, the founder of chip and infrastructure research firm SemiAnalysis, is targeting raising $400 million, according to a securities filing. The Information reported in February that Patel’s firm SemiAnalysis, which writes Substack’s top technology newsletter, was considering raising hundreds of millions of dollars to invest in startups, expanding its role from research and analysis on the hardware that undergirds AI to also investing in some of the technologies and companies the firm covers. Patel has stakes in 20 or so startups, including Mira Murati’s Thinking Machines Lab and chip startup Enfabrica, The Information reported in April. Patel previously raised a $50 million special purpose vehicle to invest in infrastructure startup Fluidstack, The Information also reported. He joins a swelling coterie of tech personalities who have parlayed fandom into fundraising. Harry Stebbings raised money for his fund 20VC after his podcast interviews with other venture capitalists took off. Lulu Cheng Meservey, a corporate communications strategist who has worked with clients including fintech Ramp and AI firm Safe Superintelligence, raised a $40 million VC fund in December 2025. Patel didn’t respond to multiple requests for comment.
|
|
|
|
Chinese tech giant Alibaba Group made its new flagship model, Qwen3.8-Max, widely available on Monday through application programming interfaces at lower prices than its Chinese competitor Moonshot AI’s popular Kimi K3 model. Alibaba will also make the open-source model available for download next week. The API for Qwen3.8-Max, Alibaba’s largest-ever model with 2.4 trillion parameters, costs $2 per 1 million input tokens and $6 per 1 million output tokens. By comparison, Kimi K3, Moonshot’s powerful 2.8 trillion-parameter open-source model that became a global sensation last month, charges $3 per 1 million input tokens and $15 for 1 million output tokens. Alibaba’s new model and its pricing further intensify the competition among Chinese AI developers that are racing to launch larger and more powerful open-source models in a challenge to Anthropic and OpenAI. Qwen3.8 Max now ranks No. 4 on the Frontend Code Arena coding leaderboard, behind two versions of Claude Opus 5 and Kimi K3.
|
|
|
|
| |
