# webAI TwIL-8B: an 8.8B reasoning model that reports 98.2 on MATH-500, non-commercial weights

Canonical: https://theagenticweekly.com/news/twil-8b-open-weight-reasoning-model · Evidence: Reported · First seen: 2026-11-04 23:13 UTC · Source published: not stated by the source · Brief published: 2026-11-05 00:04 UTC · Updated: 2026-11-05 00:04 UTC · Author: Agentic Weekly

> webAI released TwIL-8B on November 4, 2026, an 8.79B-parameter reasoning model post-trained from IBM's Granite 4.2 8B that its model card says scores 98.2 on MATH-500. The weights are on Hugging Face under the webAI Non-Commercial License, with a 4.98 GiB Q4_K_M GGUF for local use.

Models in this brief: [TwIL-8B](https://theagenticweekly.com/models/twil-8b)

## What changed

- New open-weight model: an 8.79B-parameter dense decoder-only model post-trained from ibm-granite/granite-4.2-8b, with a 131,072-token context window inherited from the base model. Text in, text out, English only.
- Two variants share one architecture and tokenizer: balanced (the default, shorter reasoning traces) and verbose (longer reasoning for hard contest problems). Both ship as full merged models, not LoRA adapters.
- webAI reports 98.2 on MATH-500, 88.3 on AIME 2025 and 81.6 on LiveCodeBench (verbose variant), the top scores in its comparison against Qwen3.5-9B, Gemma 4 12B and Gemma 4 31B.
- Knowledge is the weak spot by the card's own numbers: GPQA Diamond is 70.2, 11.5 points behind Qwen3.5-9B and 14.1 behind Gemma 4 31B.
- The card reports about 19.5k output tokens per AIME 2026 problem, 38% fewer than Qwen3.5-9B's 31.2k, at 4.2 points lower accuracy.
- Weights are 16.38 GiB per variant in bf16, with GGUF builds for llama.cpp down to a 4.98 GiB Q4_K_M. The license is the webAI Non-Commercial License ver. 1.0.

## In the source's words

> "It is level with Qwen3.5-9B on contest math while using fewer tokens per problem."
> (webAI, TwIL-8B model card on Hugging Face, https://huggingface.co/webAI-Official/TwIL-8B)

> "A short budget truncates the reasoning and scores far worse."
> (webAI, TwIL-8B model card on Hugging Face, https://huggingface.co/webAI-Official/TwIL-8B)

## What we have not verified

- We did not run TwIL-8B; every TwIL-8B score is webAI's own internal high-effort pass@1 run, and competitor figures were copied from other published sources, not re-run.
- The card says the LiveCodeBench version and date window for TwIL-8B's run were not recorded, while competitor scores are v6, so that gap is not like for like.
- Rows marked as verbose-only (LiveCodeBench, MMLU-Pro, HLE, IFBench and others) have not been run on the default balanced variant.
- The model card has no publish timestamp; the November 4, 2026 date comes from the card's last update and press coverage the same day.

## Timeline

- Source published: not stated by the source
- First seen by our watcher: 2026-11-04 23:13 UTC (via MarkTechPost coverage (signal); primary source is webAI's Hugging Face model card)
- Brief published: 2026-11-05 00:04 UTC

## Sources

- [Hugging Face model card: webAI-Official/TwIL-8B](https://huggingface.co/webAI-Official/TwIL-8B) (primary source) (fetched 2026-11-04)

## Related

- [TwIL-8B: specs, price and alternatives](https://theagenticweekly.com/models/twil-8b)
- [All news briefs](https://theagenticweekly.com/news)
- [This week's releases](https://theagenticweekly.com/this-week)
- [New AI models this week](https://theagenticweekly.com/models/this-week)
- [AI model releases by month](https://theagenticweekly.com/models/releases)
- [Best LLM for coding](https://theagenticweekly.com/best/llm-for-coding)

## How to cite

Agentic Weekly, "webAI TwIL-8B: an 8.8B reasoning model that reports 98.2 on MATH-500, non-commercial weights", published 2026-11-05 00:04 UTC, https://theagenticweekly.com/news/twil-8b-open-weight-reasoning-model