| |||
|
Third Eye
AI news straight from the labs. Twice a week, in five minutes.
| |||
| |||
|
This issue — OpenAI says it is working to make ChatGPT better at low-resource programming languages, with new evaluations covering software engineering, SQLite and COBOL migration, while Anthropic is committing 5% of its shares to a new Public Benefit Fund. In China, the theme this week was fast, cheaper models: MiniMax launched M3.1 Flash for agentic and coding work, and DeepSeek released V4.1-Flash-1017 with open weights. | |||
|
Top story · OpenAI · US · Oct 19
OpenAI targets ChatGPT coding gains in low-resource programming languages
OpenAI says it is working to improve ChatGPT's performance in low-resource programming languages. The announcement introduces new evaluations covering software engineering, SQLite, and COBOL migration.
Why it matters — Teams maintaining legacy or less common codebases, such as COBOL systems, may see OpenAI measure and improve ChatGPT's coding support in those languages, though the announcement excerpt gives no results or availability details. Read the OpenAI post → | |||
|
United States
Anthropic · US · Oct 18
Anthropic commits 5% of its shares to a new Public Benefit Fund, with more tied to valuation growth
Anthropic is creating the Anthropic Public Benefit Fund, a wholly-owned LLC that, subject to stockholder approval, will receive 5% of the company's shares, with an additional 2% possible at the board's discretion for each 50% rise in company value above a baseline. The Fund is meant to allow giving outside the company's core operations, such as large donations to nonprofits or direct distributions to specific populations, with labor market impacts of AI cited as one focus. Anthropic says details on governance and the valuation milestones will follow in the coming months.
Why it matters — A material and growing slice of a leading AI lab's equity is being set aside for public-benefit spending, including potential support for workers affected by AI, and is barred from use in lobbying or political giving. Read the Anthropic post → | |||
|
China
MiniMax · China · Oct 19
MiniMax releases M3.1 Flash, citing faster inference and 70% lower agentic task costs
MiniMax has launched M3.1 Flash, a model aimed at agentic tasks, coding, computer use and cybersecurity; an early version ran anonymously as "Space Bunny." The company reports benchmark results including 73.7% on DeepSWE v1.1 and 82% on CyberGym, and says more aggressive sparsity makes online inference nearly three times faster. It is available now via the MiniMax API, M Plan and MiniMax Code, with weights and a Pro model to follow later.
Why it matters — Teams running agent or coding workloads get a low-priced option at $0.15/$0.60 per 1M input/output tokens that MiniMax says cuts agentic usage costs by 70% versus its predecessor, with open weights promised later. Read the MiniMax post →DeepSeek · China · Oct 17
DeepSeek releases V4.1-Flash-1017 with open weights and opens Search & Fetch APIs to all users
DeepSeek has released DeepSeek-V4.1-Flash-1017, an updated model it says scores higher on benchmarks, completes tasks faster and produces shorter outputs than the previous version. It is available on DeepSeek's website, app and API, with weights published on Hugging Face and ModelScope. DeepSeek also opened its Search & Fetch APIs to everyone at a per-call price.
Why it matters — Teams using DeepSeek for coding agents can switch to an open-weight model the company says is both cheaper to run and stronger, and can now add web search and fetch at a published per-call rate, with server-side tool calls in some agent tools free until Oct 31, 2026. Read the DeepSeek post → | |||
| |||
|
Know someone who should read this? Pass it on. Free subscriptions at wave-3.co/newsletter. | |||
|
Third Eye is published by Wave3 Labs. Every story links to its official source. Summaries are AI-assisted and reviewed by an editor before sending.
|
Published by Wave3 Labs