Pending benchmarksTracking since Nov 4, 2026
Is TwIL-8B state of the art?Pending benchmarks.Announced Nov 4, 2026; no independent results yet.
webAI's 8.8B open-weight text model, post-trained from IBM Granite 4.2 8B with full supervised fine-tuning and GRPO/DAPO reinforcement learning. Ships in balanced and verbose variants. Weights under a webAI non-commercial license.
Announced facts
The details so far
- Lab
- webAI
- Announced
- Nov 4, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- 8.79B (8,791,592,960)
- Base model
- ibm-granite/granite-4.2-8b (fine-tune, not trained from scratch)
- Post-training
- Full SFT, then GRPO/DAPO reinforcement learning
- Variants
- Balanced and verbose
- Context
- 131,072 tokens
- Languages
- English, text only
- License
- webAI Non-Commercial License v1.0 (base model is Apache-2.0)
- Quantized size
- Q4_K_M GGUF is 4.98 GiB
Tags marked unsure are not stated on the lab's own pages.
Next steps
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
Self-reported
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: 98.2 on MATH-500, 88.3 on AIME 2025 and 74.4 on IFBench.
Lab claim: 81.6 on LiveCodeBench with the verbose variant.
Receipts
Sources
LABwebAI TwIL-8B announcement on Xx.com/i/status/2118054125737070659 · Nov 4, 2026REPOwebAI-Official/TwIL-8B on Hugging Facehuggingface.co/webAI-Official/TwIL-8BTracking continues until an independent board lists it.