iFlytek · Spark X2.5
Small open-weight (Apache-2.0) iFlytek on-device model with a native 1M-token context, aimed at local agent and coding use; a 1.7B sibling was released alongside.
Leader: Claude Opus 5.5 at 89.9%
Leader: Claude Sonnet 5 at 85.2%
Leader: Claude Opus 5.5 at 66.9%
Leader: GLM-5.2 at 99.1%
Leader: GPT-6 Astra at 91.5%
Leader: Kimi K3 at 84.2%
Leader: ERNIE 5.1 at 99.6%
Leader: GLM-5.2 at 92.5%
Leader: GPT-6 Astra at 96.1%
Leader: Claude Opus 5.5 at 61.4%
Leader: Kimi K3 at 88.7%
| Result | Score | Reported by | Source |
|---|---|---|---|
| BFCL-V4 | 65.1% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| tau3-bench | 30.4% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| MCP-Mark | 14.2% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| Workspace Bench | 31.2% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| VitaBench 2.0 | 25.2% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| SWE-bench Multilingual | 53.3% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| IMO-AnswerBench | 74.2% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| IFEval | 93% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| IFBench | 75% | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |
| Gaokao 2026 (5 papers) | 133.4 points out of 150 | vendor | Hugging Face: XHToken/Spark-X2.5-4B (model card) ↗ |