# Verification on VerifyBench Hard 1.0 (test) Source: [Verification on VerifyBench Hard 1.0 (test)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:52:28.089Z Evaluation of model performance on the verification task using the test split of the VerifyBench Hard 1.0 dataset, measured by mean@3 accuracy. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Verification | | Dataset (including any stated split) | VerifyBench Hard 1.0 (test) | | Metric | Mean@3 Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 18 of 18 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Mean@3 Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Mean@3 Accuracy | Result date | Report paper | | --- | --- | --- | --- | | CoSineVerifier-Tool-4B | 91.9 | Not recorded | 692e72fbd6b0dd3342932618 | | CoSineVerifier-32B | 90 | Not recorded | 692e72fbd6b0dd3342932618 | | CompassVerifier-32B | 88.9 | Not recorded | 692e72fbd6b0dd3342932618 | | o3 | 88.7 | Not recorded | 692e72fbd6b0dd3342932618 | | Qwen3-235B-A22B-2507 | 87.7 | Not recorded | 692e72fbd6b0dd3342932618 | | CompassVerifier-32B (CoT) | 86.5 | Not recorded | 692e72fbd6b0dd3342932618 | | Gemini2.5-Flash | 86 | Not recorded | 692e72fbd6b0dd3342932618 | | CoSineVerifier-4B | 85.4 | Not recorded | 692e72fbd6b0dd3342932618 | | Qwen3-32B | 85.2 | Not recorded | 692e72fbd6b0dd3342932618 | | CompassVerifier-7B | 85.2 | Not recorded | 692e72fbd6b0dd3342932618 | | GPT-oss-20B | 84.7 | Not recorded | 692e72fbd6b0dd3342932618 | | GPT-4o | 84.6 | Not recorded | 692e72fbd6b0dd3342932618 | | Qwen3-4B-Instruct-2507 | 84.1 | Not recorded | 692e72fbd6b0dd3342932618 | | Qwen3-8B | 83.6 | Not recorded | 692e72fbd6b0dd3342932618 | | XVerify-8B-I | 83.3 | Not recorded | 692e72fbd6b0dd3342932618 | | CompassVerifier-7B (CoT) | 82.6 | Not recorded | 692e72fbd6b0dd3342932618 | | Qwen3-4B | 80.3 | Not recorded | 692e72fbd6b0dd3342932618 | | LLaMA3.3-70B-Instruct | 77.2 | Not recorded | 692e72fbd6b0dd3342932618 | ### CoSineVerifier-Tool-4B Mean@3 Accuracy: 91.9. Result ID: 699998be5698665ec40d159c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### CoSineVerifier-32B Mean@3 Accuracy: 90. Result ID: 699998be5698665ec40d15a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"Labeling Verifier"}. This record does not assert independent reproduction. ### CompassVerifier-32B Mean@3 Accuracy: 88.9. Result ID: 699998be5698665ec40d159f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"Labeling Verifier"}. This record does not assert independent reproduction. ### o3 Mean@3 Accuracy: 88.7. Result ID: 699998be5698665ec40d1590. [Reporting paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Closed-source"}. This record does not assert independent reproduction. ### Qwen3-235B-A22B-2507 Mean@3 Accuracy: 87.7. Result ID: 699998be5698665ec40d1599. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### CompassVerifier-32B (CoT) Mean@3 Accuracy: 86.5. Result ID: 699998be5698665ec40d159b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### Gemini2.5-Flash Mean@3 Accuracy: 86. Result ID: 699998be5698665ec40d1592. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Closed-source"}. This record does not assert independent reproduction. ### CoSineVerifier-4B Mean@3 Accuracy: 85.4. Result ID: 699998be5698665ec40d15a0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"Labeling Verifier"}. This record does not assert independent reproduction. ### Qwen3-32B Mean@3 Accuracy: 85.2. Result ID: 699998be5698665ec40d1597. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### CompassVerifier-7B Mean@3 Accuracy: 85.2. Result ID: 699998be5698665ec40d159e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"Labeling Verifier"}. This record does not assert independent reproduction. ### GPT-oss-20B Mean@3 Accuracy: 84.7. Result ID: 699998be5698665ec40d1593. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### GPT-4o Mean@3 Accuracy: 84.6. Result ID: 699998be5698665ec40d1591. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Closed-source"}. This record does not assert independent reproduction. ### Qwen3-4B-Instruct-2507 Mean@3 Accuracy: 84.1. Result ID: 699998be5698665ec40d1598. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### Qwen3-8B Mean@3 Accuracy: 83.6. Result ID: 699998be5698665ec40d1596. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### XVerify-8B-I Mean@3 Accuracy: 83.3. Result ID: 699998be5698665ec40d159d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"Labeling Verifier"}. This record does not assert independent reproduction. ### CompassVerifier-7B (CoT) Mean@3 Accuracy: 82.6. Result ID: 699998be5698665ec40d159a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### Qwen3-4B Mean@3 Accuracy: 80.3. Result ID: 699998be5698665ec40d1595. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. ### LLaMA3.3-70B-Instruct Mean@3 Accuracy: 77.2. Result ID: 699998be5698665ec40d1594. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Verifier Type":"CoT Verifier","Model Access Category":"Open-source"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.