AuraTracer智迹闻
中文

EVENT DOSSIER

短短一周内,除了OpenAI,Anthropic、谷歌等头部厂商密集发布新品。

2026-09-04 10:24 Models 🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
5mentions
SummaryAI generated

In the early hours of September 4, 2026, OpenAI officially released its new generation flagship model, GPT-6 Astra. Greg Brockman, the company’s president, clearly stated in a announcement to the developer community that this model marked “the era of AGI”. In terms of performance, GPT-6 Astra scored 64.6% on the mathematical ability benchmark Terminal-BenchScience0.1, outperforming Claude Fable5.1’s 52.6%; it achieved 57.9% on the programming ability benchmark Terminal-Bench4.0, nearly twice as high as GPT-5.6Sol. In the OSWorld automation task, the average time taken to complete individual tasks decreased from 75 minutes to 40 minutes, a 47% improvement in efficiency. Meanwhile, within just one week, other leading companies such as Anthropic and Google also released their new models.

Related eventsRELATED EVENTS
Key entitiesKEY ENTITIES
AnthropicClaude Fable5.1GPT-6 AstraGoogleOpenAI

Coverage · reports per dayLANGUAGE SPLIT

Entity relations
Anthropic × Claude Fabl…1Anthropic × GPT-6 Astra1Anthropic × Google1Anthropic × OpenAI1Claude Fable5.1 × GPT-6…1Claude Fable5.1 × Google1

SignalsSIGNALS

Keyword heat
  • OpenAI1
  • Anthropic1
  • Google1
  • GPT-6 Astra1
  • Claude Fable5.11

All reports (1)SOURCES

界面新闻·科技早报 en 2026-09-04 10:24

In just one week, besides OpenAI, leading companies such as Anthropic and Google have released new products in quick succession.

# In just one week, leading companies like OpenAI, Anthropic, and Google have released new products in bulk. ## Key Events In the early morning of September 4th, OpenAI officially launched its new generation flagship model **GPT-6 Astra**. Company President Greg Brockman said in a announcement to the developer community, “Welcome to the AGI era.” ## Product Highlights | Dimension | Performance | |----------|------------| | **Mathematical Capability** | Scored 64.6% on the Terminal-BenchScience0.1 benchmark, outperforming Claude Fable5.1’s 52.6% | | **Programming Ability** | Scored 57.9% on Terminal-Bench4.0, nearly doubling compared to GPT-5.6Sol | | **OSWorld Automation** | Average time for completing individual tasks decreased from 75 minutes to 40 minutes, with an efficiency improvement of 47% | | **Mathematical Reasoning**…