U.S. companies say they are increasing their use of open-weight models
BIBIBI
AT A GLANCE
The report says that U.S. companies are increasingly using open-weight models to handle simple AI tasks in response to rising costs.
Article
Major U.S. companies are collectively beginning to use cheaper AI: mentions of open-source models rose year-over-year by 6 times
Beating News Flash,AI reports that a Financial Times survey found that open-weight models are spreading from technology companies to traditional U.S. businesses. AlphaSense data show that this year August and September, the number of times U.S. companies mentioned open-weight or open-source models during earnings calls and investor conferences increased by approximately 6 times year-over-year.PNC Financial, logistics company CH Robinson, and Siemens have recently publicly discussed related applications.
Companies are beginning to switch models, with cost being the most direct reason. Tinder's AI annualized spending rose from January approximately 1000000 USD to July approximately 10000000 USD, and it has now begun routing some routine requests to open-weight models.AT&T operates at a larger scale. Currently, approximately 40%of its AI tasks have been assigned to open models, and the company plans to further increase this to approximately 70%. Its internal AI system processes approximately 45000000000 Token per day; simple tasks are assigned to cheaper models, while complex tasks continue to use frontier models such as OpenAI and Anthropic.
Original link https://m.theblockbeats.info/flash/369301
Key points
01
A Financial Times survey said that open-weight models are spreading from technology companies to traditional U.S. businesses.
02
AlphaSense data show that this year August and September, the number of times U.S. companies mentioned open-weight or open-source models during earnings calls and investor conferences increased by approximately 6 times year-over-year.
03
PNC Financial、CH Robinson and Siemens recently publicly discussed related applications.
04
Tinder's AI annualized spending rose from January approximately 1000000 USD to July approximately 10000000 USD, and it began routing some routine requests to open-weight models.
05
AT&T currently assigns approximately 40%of its AI tasks to open models and plans to increase the share to approximately 70%。
06
AT&T's internal AI system processes approximately 45000000000 Token per day; simple tasks use cheaper models, while complex tasks continue to use frontier models such as OpenAI and Anthropic.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Open-weight models allow companies to deploy and use models more flexibly. The approach described in the report allocates models according to task difficulty: routine tasks are assigned to lower-cost models, while complex tasks continue to use frontier models, in an effort to control AI spending.
Why it matters to readers
If more companies adopt this division-of-labor approach, they may place greater emphasis on cost when selecting AI models and allocate different tasks between open models and frontier models. The report also indicates that this does not mean frontier models are being comprehensively replaced.
Beginners can understand this as companies no longer using the same high-cost model for all AI tasks, but instead selecting different models according to task difficulty.
Risks and unknowns
This news brief does not include the Financial Times' original investigation, AlphaSense's statistical details, or the statistical methodology.
The corporate spending, task share,Token processing volume, and future plans mentioned in the article have not been independently verified through corporate announcements or original records.
“An increase in mentions” does not equate to an increase in actual deployment scale.
The report does not explain the actual effects of open models in terms of cost, performance, security, and compliance.
Related Developments
Loading event timeline…
Related concepts
Open-weight models
This term is not in the glossary yet. Browse related concepts in the glossary.
