Mercury 2.5 LLM achieves 770 tokens per second generation speed

·1 views

Mercury 2.5 is a large language model that has reached a generation speed of 770 tokens per second, according to data published on Artificial Analysis. This figure places it among the faster models currently being benchmarked on the platform. The milestone was noted by the Hacker News community, where the announcement attracted early discussion. Speed benchmarks like this are increasingly relevant as developers evaluate LLMs for latency-sensitive applications.

Read the full story at Hacker News

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

NIGHTCELL 7 adds second Viking metal-hip-hop album built entirely with Python code

NIGHTCELL 7, a free browser and desktop sandbox game, has released a second in-game soundtrack album titled 'When the Ravens Lied' by Þrøngva, featuring twelve new tracks blending drop-D Viking metal and hip-hop. The album retells the game's MIRAGE storyline as a Norse saga, in which forged messages from Odin's ravens manipulate two warring halls while a merchant profits from the conflict. Every note was synthesised programmatically using a Python script with numpy and ffmpeg — no samples, AI audio tools, or music services were used. Because the creator cannot hear, the mix was validated through spectrograms and energy analysis, revealing a bass-heavy imbalance that was corrected before final rendering at 320 kbps. Both albums are available free at nightcell7.com/play, streaming on demand, with lyrics embedded in each MP3 and the full synthesiser code published on GitHub.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

ChatGPT Surpasses 1 Billion Weekly Users, Serving 2.5 Million Businesses

OpenAI announced on September 8, 2026, that its ChatGPT product family now exceeds one billion weekly active users and serves 2.5 million businesses worldwide. The figures cover OpenAI's broader product suite rather than any single ChatGPT tier or region. While the milestone signals widespread employee and consumer familiarity with AI assistants, it does not indicate that every organization has a mature or effective AI strategy in place. Experts note the more pressing question for businesses is no longer whether staff will encounter AI tools, but where those tools should and should not be integrated into real workflows. The scale does lower the barrier to experimentation, though deploying AI in systems connected to customer data or operational infrastructure requires careful governance around access, data handling, and oversight.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Two Contentful builds reveal key lessons on content modeling and editorial control

A development team implemented Contentful-based CMS solutions for two clients — nonprofit Addiction Education Society (AES) and AI platform CrewAI — each with distinct editorial needs. For AES, structured content models for resources, stories, and core pages enabled staff to publish updates in minutes instead of days, without relying on developers. CrewAI's solution used reusable Figma-designed components mapped to Contentful models, letting marketing assemble pages while engineers controlled rendering via GraphQL. AES retained its existing WordPress learning system alongside the new Astro-and-Contentful marketing platform to avoid downtime, with a migration path planned for later. The projects highlighted three critical pre-migration questions: what content is reusable, which system owns each publication step, and which existing tools should remain in place.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Run n8n and a Node.js API Locally Using Docker Compose

A tutorial published on DEV Community walks developers through building a self-hosted workflow automation environment using n8n and a custom Node.js REST API. The setup uses Docker Compose to run both services within a shared internal bridge network, avoiding exposure to the public internet. This approach addresses common concerns around latency, rate limits, and data privacy associated with cloud-based automation tools. A lightweight Express API is containerized using a Node Alpine Docker image, while n8n handles workflow orchestration and communicates with the API via Docker's internal DNS. The configuration also includes a persistent Docker volume to ensure n8n workflows are retained even when containers are stopped or rebuilt.

0 comments Read more at DEV Community