<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/gDlTrnG6P" -->

---
title: webAI's 8.8B open-weight model claims 93% of frontier...
description: webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware. Its weights are available on Hugging Face. The model...
canonical: https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: webAI's 8.8B open-weight model claims 93% of frontier coding performance | daily.dev
og:description: webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware. Its weights are available on Hugging Face. The model...
og:url: https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p
og:image: https://api.daily.dev/og/posts/gDlTrnG6P.png
og:image:alt: webAI's 8.8B open-weight model claims 93% of frontier coding performance
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# webAI's 8.8B open-weight model claims 93% of frontier coding performance

**[Collections](https://daily.dev/sources/collections)** · 2 min read · 15 upvotes · 4 comments

## Summary

webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware. Its weights are available on Hugging Face. The model reportedly reaches 93% of frontier performance on coding and 92% on instruction following, with @Hesamation citing about 90% on contest math. These are relative figures, so the benchmarks and frontier models used for comparison matter before trusting the percentages. If the numbers hold, a local model that gets close to frontier quality on coding could cut reliance on expensive API calls.

## Content

webAI has released TwIL-8B, an 8.79B-parameter reasoning model post-trained from IBM's Granite 4.2 8B. It targets math, coding and instruction following, and the weights are on Hugging Face under a non-commercial license.

## Size and hardware

The Q4_K_M GGUF build is 4.98 GiB, small enough to run on a CPU or a modest graphics card. That's the pitch: near-frontier coding and math on hardware people already own, and fewer expensive API calls.

## Benchmarks

These are webAI's own comparisons, so treat them accordingly.

- **MATH-500:** 98.2, the highest score in webAI's comparison.
- **LiveCodeBench:** 81.6 for the verbose variant. That is 16.0 points above Qwen3.5-9B and ahead of Google's Gemma 4 31B, a model about 3.5x its size.
- **Against frontier models:** compared with GPT-5.5, Claude Opus 4.8 and Grok 4.3, TwIL-8B reaches 93% of the best score on coding and 92% on instruction following. On IFBench it beats Claude Opus 4.8 by 12.2 points. One summary also puts it at about 90% of frontier contest-math performance.
- **Token use:** a final training stage rewards shorter reasoning. On AIME 2026 the model uses 38% fewer output tokens than Qwen3.5-9B.

## Where it falls short

Knowledge benchmarks are weaker. TwIL-8B scores 70.2 on GPQA Diamond, against 81.7 for Qwen3.5-9B. The non-commercial license also rules out use in products.

## Reaction

@svpino wrote that open models keep getting better and that @thewebAI "cooked here." @Hesamation and @rohanpaul_ai made similar points about running near-frontier coding locally. I'd want independent runs of these numbers before leaning on them, but an 8.8B model that fits in under 5 GiB and posts these results is worth trying.

## Questions this post answers

### What is webAI's 8.8B parameter open-weight model and how does it perform on coding benchmarks?

webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware, with weights published on Hugging Face. It reportedly reaches 93% of frontier performance on coding and 92% on instruction following, while one reviewer separately measured around 90% on contest math benchmarks against frontier models.

_Developers weighing local models against frontier APIs can track releases like this on daily.dev._

## Community take

How the wider developer community reacted, aggregated from 1 discussion and 12 comments across x (as of 2026-11-06).

**TL;DR:** Developers are excited that an 8.8B model fits under 5GB and runs on a CPU or modest GPU, but many want proof beyond benchmark charts before accepting the near-frontier coding claim.

**Sentiment:** 45% positive · 45% mixed · 10% skeptical

**The case for**

- Fitting a near-frontier coding model under 5GiB quantized makes local/offline use realistic.
- Lower token usage versus a similarly sized rival could make it cheaper to run as an agent than chasing a one-off frontier score.
- Running without a cluster or datacenter is seen as more broadly useful than raw frontier scores.

**The pushback**

- The frontier comparison set is a generation behind on some models, so the 'near-frontier' framing feels premature.
- Unclear whether benchmark scores were measured on the quantized build or the full bf16 weights.
- Some want the exact LiveCodeBench split and real-world coding tests rather than just aggregate benchmark numbers.

**By community**

- x (mixed): Replies welcome the small, CPU-friendly footprint but repeatedly question whether the benchmarks hold up under scrutiny or real coding use.

**Hottest debate:** Whether comparing against a frontier set that's a generation behind two of three models invalidates the 'near-frontier' claim.

**Open questions**

- Were the reported benchmark scores run on the quantized weights or the full bf16 model?
- How does it perform on long, multi-step coding tasks compared to larger open models?
- What is the exact LiveCodeBench split used to produce the 81.6 score?

**Highlights**

> @rohanpaul_ai @thewebAI That frontier set is GPT-5.5 and Opus 4.8, so it's a generation behind on two of the three. I'd like to see the same chart against GPT-6 and Opus 5.5 before calling it near-frontier.
> — [AIQuanting on x](https://x.com/AIQuanting/status/2118426331092554230)

> @rohanpaul_ai @thewebAI 4.98 GiB for 8.8B params works out to about 4.9 bits per weight. was the 81.6% LiveCodeBench run on that quant or on the full bf16 weights?
> — [Chahatusharma on x](https://x.com/Chahatusharma/status/2118427687131742564)

> @rohanpaul_ai @thewebAI the 5gb quantized file is the real story. near-frontier coding that fits a modest mac beats a 1.5tb open weight that needs a server rack
> — [rudraxx13 on x](https://x.com/rudraxx13/status/2118429924348313947)

> @rohanpaul_ai @thewebAI 38% fewer tokens vs Qwen3.5-9B on AIME is the agent line. a 5 GB local model that finishes cheaper beats a frontier score you only hit once
> — [martinOlsen94M on x](https://x.com/martinOlsen94M/status/2118424783683772801)

> @rohanpaul_ai @thewebAI An 8B open model tuned for math and code is a size people can actually run locally. Curious how it holds up on long multi-step coding tasks vs bigger open models.
> — [Righteousbeing7 on x](https://x.com/Righteousbeing7/status/2118453201121018112)

**Source threads**

- [x](https://x.com/rohanpaul_ai/status/2118424398642532579) · 0 points · 12 comments

## Community discussion

Top comments from developers on daily.dev.

**@tekn0wledg** · 5 upvotes

> Hmm. I suspect hype from webAI, but gonna bench it regardless. Could be interesting.

**@vapeanddebug** · 0 upvotes

> good model, just verbose, even the balanced version

**@pedropietro** · 0 upvotes

> ![GIF](https://static.klipy.com/ii/da290b156d64898341638f3c299e7478/57/14/vdgs2tLt.gif)

**@sayanmondal** · 0 upvotes

> ![GIF](https://static.klipy.com/ii/35ccce3d852f7995dd2da910f2abd795/ef/e8/lWzgchDM.gif)

## Similar posts on daily.dev

- [ThursdAI - Jul 16 - Inkling 975B open weights, Kimi K3 at 2.8T, a 27B model on a phone & Codex hits 9M](https://daily.dev/posts/thursdai---jul-16---inkling-975b-open-weights-kimi-k3-at-2-8t-a-27b-model-on-a-phone-codex-hits--hledq9pdh) · ThursdAI · 0 upvotes · 0 comments

---

Tags: [#ai](https://daily.dev/tags/ai), [#open-source](https://daily.dev/tags/open-source), [#ai-coding](https://daily.dev/tags/ai-coding), [#huggingface](https://daily.dev/tags/huggingface)

[View this post on daily.dev](https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"webAI's 8.8B open-weight model claims 93% of frontier coding performance","url":"https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p"},"datePublished":"2026-11-04T19:30:47.559Z","dateModified":"2026-11-06T12:36:14.889Z","description":"webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware. Its weights are available on Hugging Face. The model...","image":"https://pbs.twimg.com/media/HWTfUzNXUAAsciL.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HWTfUzNXUAAsciL.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":4,"discussionUrl":"https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":15},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":4}],"keywords":"ai,open-source,ai-coding,huggingface","timeRequired":"PT2M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"webAI's 8.8B open-weight model claims 93% of frontier coding performance"}]}
{"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p","comment":[{"@type":"Comment","text":"Hmm. I suspect hype from webAI, but gonna bench it regardless. Could be interesting.","datePublished":"2026-11-05T01:33:52.387Z","url":"https://daily.dev/posts/gDlTrnG6P#c-tfFNqjZ9T","author":{"@type":"Person","name":"Scott Davis","url":"https://daily.dev/tekn0wledg","image":"https://avatars.githubusercontent.com/u/355212?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5}},{"@type":"Comment","text":"good model, just verbose, even the balanced version","datePublished":"2026-11-06T04:28:26.302Z","url":"https://daily.dev/posts/gDlTrnG6P#c-R4iwVIBvh","author":{"@type":"Person","name":"Clarence Middleton","url":"https://daily.dev/vapeanddebug","image":"https://media.daily.dev/image/upload/s--O0TOmw4y--/f_auto/v1715772965/public/noProfile"}},{"@type":"Comment","text":"","datePublished":"2026-11-06T12:42:05.542Z","url":"https://daily.dev/posts/gDlTrnG6P#c-63UBBjbMx","author":{"@type":"Person","name":"Pedro Pietroluongo","url":"https://daily.dev/pedropietro","image":"https://avatars.githubusercontent.com/u/1533471?v=4"}},{"@type":"Comment","text":"","datePublished":"2026-11-06T15:39:18.057Z","url":"https://daily.dev/posts/gDlTrnG6P#c-QqcC7zhq0","author":{"@type":"Person","name":"Sayan Mondal","url":"https://daily.dev/sayanmondal","image":"https://media.daily.dev/image/upload/s--cW6TLpb9--/f_auto/v1731954927/avatars/avatar_31X8v4aef86WyygmnVrjC"}}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/webai-s-8-8b-open-weight-model-claims-93-of-frontier-coding-performance-gdltrng6p#faq","mainEntity":[{"@type":"Question","name":"What is webAI's 8.8B parameter open-weight model and how does it perform on coding benchmarks?","acceptedAnswer":{"@type":"Answer","text":"webAI released an open-weight model with 8.8 billion parameters, small enough to run on local hardware, with weights published on Hugging Face. It reportedly reaches 93% of frontier performance on coding and 92% on instruction following, while one reviewer separately measured around 90% on contest math benchmarks against frontier models. Developers weighing local models against frontier APIs can track releases like this on daily.dev."}}]}
```

