- Lenovo TianxiAI's TianxiCode code agent framework scored 71% on SWE-bench-Live Lite, ranking first globally with official Verified certification.
- The system uses DeepSeek-v4.1-Flash as its foundation model, with TianxiCode providing the agent architecture layer for code navigation, planning and self-correction.
- Lenovo says TianxiCode will be integrated into its AI hardware product lineup, targeting real-world developer toolchains rather than benchmark performance alone.
Global First on SWE-bench-Live
Lenovo TianxiAI announced that its self-developed code agent framework TianxiCode achieved a 71% problem-solving rate on SWE-bench-Live (Lite), an internationally recognized benchmark that evaluates AI systems on real-world GitHub issue resolution. The result earned TianxiCode the top global ranking and official Verified certification from the benchmark administrators, Lenovo said.
SWE-bench-Live differs from simpler code-generation tests by requiring systems to understand existing codebases, locate bugs across files and produce working patches in reproducible execution environments. The Verified designation means the submission passed an audit of agent run trajectories to confirm no leakage of test answers or results.
Agent Architecture Over Raw Model Power
Lenovo positioned the achievement as evidence that agent architecture matters as much as the underlying model. The company described DeepSeek-v4.1-Flash as the reasoning engine while TianxiCode provides the engineering workflow, including cross-file multi-hop retrieval, autonomous planning with multi-turn tool calling, and closed-loop patch generation with test-driven self-correction.
According to Lenovo, these capabilities allow the system to navigate large codebases, run diagnostic commands, and iteratively fix its own output when tests fail, mimicking how a human engineer approaches debugging.
From Benchmark to Hardware Integration
Lenovo stated that TianxiCode is a core component of its TianxiAI ecosystem's code intelligence capabilities and will be deployed across the company's AI hardware products. The goal, Lenovo said, is to reduce developers' debugging and engineering coordination costs by embedding autonomous code-repair tools directly into device-level toolchains.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
