webAI Releases TwIL-8B, an 8.8B Reasoning Model Scoring 98.2 on MATH-500

webAI has released TwIL-8B, an 8.8B reasoning model for math, code and instruction following. The TwIL-8B reasoning model is post-trained from IBM’s Granite 4.2 8B. Weights are on Hugging Face under a non-commercial license. webAI’s launch post frames the pitch around cost per task, not raw scale. The 4-bit build is 4.98 GiB, sized for CPU or modest-VRAM GPU inference.

TL;DR

• Size: 8.79B parameters, dense. 131,072-token context, inherited from the base model.
• Runs on: CPU or modest-VRAM GPU via a 4.98 GiB Q4_K_M GGUF. Full bf16 weights are 16.38 GiB.
• Performance: Leads its size class on math and code, trails on knowledge benchmarks.
• Best: MATH-500 at 98.2, the top score in webAI’s comparison.
• Worst: GPQA Diamond at 70.2, 11.5 points behind Qwen3.5-9B. HLE is 12.6 (verbose variant).

What is TwIL-8B?

TwIL-8B is a dense decoder-only reasoning model from webAI. It has 40 layers, hidden size 4096, and 32 attention heads with 8 KV heads. Input and output are English text only. The model writes a block before each answer.

It ships in 2 variants with the same architecture and tokenizer. Balanced is the default and produces shorter reasoning traces. Verbose reasons longer and targets hard contest-style problems.

TwIL-8B extends webAI’s TwIL family. The first members were 1.7B and 3B formal-logic models released in August 2026.

How was TwIL-8B trained?

webAI used 3 post-training stages:

1. Full-parameter SFT on reasoning traces for math, STEM and instruction following.
2. GRPO/DAPO reinforcement learning on math, code, STEM and instruction prompts. This produces the verbose variant.
3. Length-aware RL with a reward that favors shorter traces against per-domain length references. This produces the balanced variant.

Training pools were decontaminated with 13-gram matching against AIME, HMMT, MATH-500, GPQA and other evals.

How does TwIL-8B perform on benchmarks?

All TwIL-8B scores are self-reported, high-effort pass@1 with thinking enabled. Scores marked † were measured on the verbose variant only.

• Math: TwIL-8B scores 98.2 on MATH-500. Qwen3.5-9B lists 96.6 on a third-party leaderboard. Gemma 4 31B scored 87.9 in Red Hat’s lighteval run. On AIME 2025, TwIL-8B posts 88.3 against 87.1 for Gemma 4 31B. On AIME 2026 it scores 88.3, behind Qwen3.5-9B at 92.5 and Gemma 4 31B at 89.2.
• Code: LiveCodeBench† is 81.6, against 65.6 for Qwen3.5-9B and 80.0 for Gemma 4 31B. webAI notes its LiveCodeBench version was not recorded. Competitor scores are v6.
• Instruction following: IFBench† is 74.4, 9.9 points above Qwen3.5-9B at 64.5. IFEval is level, at 91.8 vs 91.5.
• Knowledge: This is the weak spot. GPQA Diamond is 70.2, against 81.7 for Qwen3.5-9B and 84.3 for Gemma 4 31B. MMLU-Pro† is 76.5 vs 82.5 for Qwen3.5-9B.

Across 13 shared benchmarks, TwIL-8B goes 7-6 against Qwen3.5-9B. Against Gemma 4 31B, about 3.5x its size, it goes 3-11.

Does TwIL-8B use fewer reasoning tokens?

Yes, against its closest open peer. On AIME 2026, TwIL-8B averages about 19.5k output tokens per problem. Qwen3.5-9B uses 31.2k, so TwIL-8B uses 38% fewer. It scores 4.2 points lower. On HMMT Feb 2026 the gap is 32.9k vs 40.6k, or 19% fewer.

Frontier models remain far leaner. webAI cites about 5k tokens per AIME 2026 problem for GPT-5.5. webAI states that harnesses and tokenizers differ, so treat this as directional.

TwIL-8B vs Qwen3.5-9B vs Gemma 4

TwIL-8B figures from the model card. † = verbose variant. All scores self-reported by each vendor unless marked.