# Tool Call Repair on Tool-Reflection-Bench (test) Source: [Tool Call Repair on Tool-Reflection-Bench (test)]() Collection index: [llms.txt]() Data updated: 2026-04-16T07:03:07.337Z Evaluation of model performance on Tool Call Repair capabilities using the test split of the Tool-Reflection-Bench dataset, reporting Repair@K metrics. 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 | Tool Call Repair | | Dataset (including any stated split) | Tool-Reflection-Bench (test) | | Metric | Repair@1 (%) | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 9 of 9 | | 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 - [Repair@1 (%)](): Higher is better; unit: PERCENT. - [Repair@3 (%)](): Higher is better; unit: PERCENT. - [Repair@5 (%)](): Higher is better; unit: PERCENT. ## Recorded results | Method | Repair@1 (%) | Result date | Report paper | | --- | --- | --- | --- | | structured reflection | 14.9 | Not recorded | 68fa7b971e209131cea89533 | | Qwen3-4B-Instruct | 9.6 | Not recorded | 68fa7b971e209131cea89533 | | structured reflection | 9.3 | Not recorded | 68fa7b971e209131cea89533 | | GPT-4o-mini | 6.1 | Not recorded | 68fa7b971e209131cea89533 | | structured reflection | 4.7 | Not recorded | 68fa7b971e209131cea89533 | | GPT-4.1-mini | 3.1 | Not recorded | 68fa7b971e209131cea89533 | | Qwen2.5-7B-Instruct | 2.4 | Not recorded | 68fa7b971e209131cea89533 | | LongCat-Lite-8K-Chat | 2.3 | Not recorded | 68fa7b971e209131cea89533 | | Llama-3.1-8B-Instruct | 0.7 | Not recorded | 68fa7b971e209131cea89533 | ### structured reflection Repair@1 (%): 14.9. Result ID: 69e0889847f34ce924c94969. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct","Model Category":"Open-Sourced Models"}. This record does not assert independent reproduction. ### Qwen3-4B-Instruct Repair@1 (%): 9.6. Result ID: 69e0889847f34ce924c94966. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Open-Sourced Models","Training":"Base"}. This record does not assert independent reproduction. ### structured reflection Repair@1 (%): 9.3. Result ID: 69e0889847f34ce924c94968. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B-Instruct","Model Category":"Open-Sourced Models"}. This record does not assert independent reproduction. ### GPT-4o-mini Repair@1 (%): 6.1. Result ID: 69e0889847f34ce924c94962. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Close-Sourced Models"}. This record does not assert independent reproduction. ### structured reflection Repair@1 (%): 4.7. Result ID: 69e0889847f34ce924c94967. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Model Category":"Open-Sourced Models"}. This record does not assert independent reproduction. ### GPT-4.1-mini Repair@1 (%): 3.1. Result ID: 69e0889847f34ce924c94963. [Reporting paper]() Recorded evaluation configuration: {"Model Category":"Close-Sourced Models"}. This record does not assert independent reproduction. ### Qwen2.5-7B-Instruct Repair@1 (%): 2.4. Result ID: 69e0889847f34ce924c94965. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Open-Sourced Models","Training":"Base"}. This record does not assert independent reproduction. ### LongCat-Lite-8K-Chat Repair@1 (%): 2.3. Result ID: 69e0889847f34ce924c94961. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Close-Sourced Models"}. This record does not assert independent reproduction. ### Llama-3.1-8B-Instruct Repair@1 (%): 0.7. Result ID: 69e0889847f34ce924c94964. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Open-Sourced Models","Training":"Base"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.