Primary use stated or inferred cautiously from official documentation.
W
WEBAI / webAI-Official/TwIL-8B
TwIL-8B
English-language reasoning model with 8.79 billion parameters, derived from IBM Granite 4.2 8B. On November 8, webAI released two full variants: balanced, the default, and verbose. Both come in BF16 and GGUF.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | webAI-Official/TwIL-8B |
|---|---|
| Model type | Text and reasoning · Code |
| Access model | Downloadable weights for non-commercial use |
| License | webAI Non-Commercial License ver. 1.0 |
| Deployment | Self-hosted |
| Release | 08 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text |
| Exit | Text · Code |
| Context window | 131,072 tokens |
| maximum output | Not published |
| Reasoning | Balanced and verbose variants |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · Transformers · llama.cpp |
Test mathematical problem-solving and code generation on your own infrastructure. Balanced uses shorter reasoning traces. The Q4_K_M quantization of each variant takes 4.98 GiB on disk.
The 131,072-token context comes from Granite. The provider has not documented a separate output limit or evaluated tool use and agent tasks. Its published tests use BF16 weights with reasoning enabled.
03 / CAPABILITIES
What it can do
Orientation
Test mathematical problem-solving and code generation on your own infrastructure. Balanced uses shorter reasoning traces. The Q4_K_M quantization of each variant takes 4.98 GiB on disk.
Context
131,072 tokens · Not published
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
Text
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Weights are available under a non-commercial license. webAI publishes no API price for this checkpoint. Self-hosting requires compute resources.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
webAI reports 98.2% on MATH-500 with balanced and 74.4% on IFBench with verbose. It also reports 81.6% on LiveCodeBench with verbose, without specifying the version or date window. Other models were tested under different protocols.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| MATH-500 | 98,2 % | pass@1 | View source ↗ |
| IFBench | 74,4 % | pass@1 | View source ↗ |
| LiveCodeBench | 81,6 % | pass@1 | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
TwIL-8B balanced
Default variant with shorter traces.
TwIL-8B verbose
Variant with longer reasoning traces.
07 / SOURCES