webAI Opens TwIL 8B Weights, an 8.8B Model It Says Nears Frontier Scores on Coding and Math
Modest intelligence is arriving faster than most people realize. Today, we're open-sourcing TwIL-8B. Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks: • 93% of frontier performance on coding • 92% on instruction following • 90% on contest mathematics All from a model small enough to run on your own hardware. But the benchmarks aren't the most important part of this announcement. For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it. I believe that equation is changing. The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task. If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model? As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own. That's the future we're building at @thewebAI TwIL-8B is available today. Open weights. And this is just the first of several announcements we're making today. HF Link:
Context
An 8.8B open weights model claiming near frontier coding and math scores pushes the cost of useful local intelligence down again.
David Stout of webAI posted that the company is open sourcing TwIL 8B, an 8.8 billion parameter model it says approaches frontier scores on coding, instruction following, and contest mathematics while running on local hardware. Weights are listed on Hugging Face. The company frames the pitch around cost per completed task rather than raw model size.
Deep dive
David Stout (@Davidstout) announced that webAI is open sourcing TwIL 8B. He describes it as 8.8 billion parameters and says it reaches about 93 percent of frontier performance on coding, 92 percent on instruction following, and 90 percent on contest mathematics, while still running on hardware people already own.
The post argues that the industry focus is shifting from building the most capable model at any cost toward the cost of completing work. If a small model can finish the same tasks as a much larger one, buyers may stop paying for the larger model by default.
Weights are pointed to Hugging Face in the announcement. webAI's own account posted a follow up the same day. Benchmark claims in the launch post should be treated as company reported until independent runs land.
This sits beside other small and midsize open weight releases on the day, including larger coding focused models, and adds a local first cost argument rather than a new frontier parameter race.
What to expect next
Watch for third party evals that confirm or cut the coding, instruction, and math claims, and for how quickly the Hugging Face weights get adapted into local agent stacks. Follow ups Stout teased as same day announcements would clarify whether TwIL 8B is a one off drop or the start of a series. Compare local task cost against other open 7B to 14B class models shipping this week.