webAILanguage modelActive

TwIL-8B

TwIL-8B has no overall index value yet because there are too few independent benchmark results.

Summary

TwIL-8B is an English-language model with 8.79 billion parameters, based on IBM Granite 4.2 8B. webAI post-trained it for math problems, coding, and instruction following.

When you load the weights locally, you get the Balanced variant by default. Verbose produces longer reasoning traces. The model card lists a 131,072-token context window. The weights use the webAI Non-Commercial License 1.0.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 7 published benchmark results

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

MATH-500 (Show measurement, test conditions, and source)

Source value: 98.2
Metric: pass@1
Unit: %
Benchmark version: MATH-500; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

AIME 2025 (Show measurement, test conditions, and source)

Source value: 88.3
Metric: pass@1
Unit: %
Benchmark version: AIME 2025; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

AIME 2026 (Show measurement, test conditions, and source)

Source value: 88.3
Metric: pass@1
Unit: %
Benchmark version: AIME 2026; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

HMMT February 2025 (Show measurement, test conditions, and source)

Source value: 85
Metric: pass@1
Unit: %
Benchmark version: HMMT February 2025; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

HMMT February 2026 (Show measurement, test conditions, and source)

Source value: 68.2
Metric: pass@1
Unit: %
Benchmark version: HMMT February 2026; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

HMMT November 2025 (Show measurement, test conditions, and source)

Source value: 86.7
Metric: pass@1
Unit: %
Benchmark version: HMMT November 2025; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

GPQA Diamond (Show measurement, test conditions, and source)

Source value: 70.2
Metric: pass@1
Unit: %
Benchmark version: GPQA Diamond; webAI release evaluation, balanced
Category: reasoning
Direction: Higher is better
Source type: vendor-reported
Evaluator: webAI
Status: active
Retrieved at: 2026-11-09
Methodology: Vendor-reported result. TwIL-8B balanced, webAI internal harness, high-effort pass@1 with thinking enabled.
Context: webAI does not specify evaluation precision or the harness revision. These scores are not directly comparable with other setups.

Source webAI

Similar models

More models of the same kind as TwIL-8B, chosen by model line, developer, size, price and release date.

All specifications

Architecture

Base model: IBM Granite 4.2 8B; Source: Base model

Input and output

Input: Text; Source: Input
Output: Text; Source: Output
Language: English; Source: Language
Reasoning: Enabled by default. Can be disabled in the chat template.; Source: Reasoning

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking: Unknown; no source
Watermarking technique: Not known; no source

Licence and access

License: webAI Non-Commercial License 1.0. Non-commercial research and education only.; Source: License
Shutdown: None announced; no source

API

Model ID: webAI-Official/TwIL-8B; Source: Model ID

Further details

Weight variants: Balanced is the default. Verbose uses longer reasoning traces. Both are available as BF16 weights and in GGUF formats.; Source: Weight variants

Frequently asked questions about TwIL-8B

How large is the context window of TwIL-8B?

131,072 tokens. That is how much text TwIL-8B reads at once, about 200 A4 pages.

Page count as a rule of thumb from Artificial Analysis

Context windowTokens the model reads at once · higher is better

GLM-5.1200K

Claude Opus 4.5200K

o3200K

TwIL-8BRank 101 of 111131K

GLM-4.5131K

GPT OSS 120B131K

DeepSeek-R1128K

Can I run TwIL-8B locally?

Yes, the weights are open.

Stated by webAI

Open weights: yes
License: webAI Non-Commercial License 1.0. Non-commercial research and education only.
Parameters: 8.8B
active per token: n/a

Sources and data date

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

Type	Evidence and data date
Input	Documented modalities. Retrieved November 9, 2026. Source: webAI TwIL-8B model card · webAI TwIL-8B model card (retrieved November 9, 2026) · webAI TwIL-8B model card · Editorial description reviewed November 9, 2026
Output	Documented modalities. Retrieved November 9, 2026. Source: webAI TwIL-8B model card · webAI TwIL-8B model card (retrieved November 9, 2026) · webAI TwIL-8B model card · Editorial description reviewed November 9, 2026
Research status	Research date November 9, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional source	webAI TwIL-8B public weights publication (retrieved November 9, 2026)
Additional source	webAI Non-Commercial License 1.0 (retrieved November 9, 2026) · webAI Non-Commercial License 1.0 · Editorial description reviewed November 9, 2026
Additional source	webAI (retrieved November 9, 2026)
Additional source	Model metadata (retrieved November 9, 2026)