AnswersPublished18 sources
Lenovo’s TianxiCode Takes First on SWE-bench-Live Lite With a 71% Score
Lenovo’s TianxiCode coding agent framework, running DeepSeek V4.1 Flash, resolved 213 of 300 SWE bench Live Lite tasks for a verified 71% score and first place on the October 8, 2026 leaderboard. The benchmark tests agents on issues drawn from real GitHub projects; the score reflects the combined model and framework...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Lenovo’s TianxiCode coding agent framework, running DeepSeek V4.1 Flash, resolved 213 of 300 SWE bench Live Lite tasks for a verified 71% score and first place on the October 8, 2026 leaderboard.
- The benchmark tests agents on issues drawn from real GitHub projects; the score reflects the combined model and framework system, not DeepSeek alone.
- Official verification included review of submitted agent trajectories and evaluation conditions intended to check for answer or test leakage.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Lenovo’s Tianxi AI team achieve with its in-house TianxiCode coding-agent framework running DeepSeek-V4.1-Flash on SWE-bench-Live L. Article summary: Lenovo’s Tianxi AI team reportedly took first place on SWE-bench-Live Lite with its in-house TianxiCode agent framework running DeepSeek-V4.1-Flash, achieving an officially verified 71% resolution rate; QbitAI reported t. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
Lenovo’s Tianxi AI team took first place on SWE-bench-Live Lite with TianxiCode, its in-house coding-agent framework, running DeepSeek-V4.1-Flash. The October 8, 2026 leaderboard entry shows a 71.0% resolution rate across 300 tasks—213 resolved—and is marked Verified. The lead over the next verified entry was small: about 0.67 percentage points. 2
16
The result is notable, but its scope matters: it is a Lite leaderboard result, not evidence that the system leads the full or multilingual boards. QbitAI reported the result on October 9, and Pandaily covered it on October 10. 5
19
44
What the 71% score measures
SWE-bench-Live evaluates whether an AI system can address software issues drawn from real GitHub projects by producing code patches. That makes it a test of issue-solving workflows, rather than only code completion or isolated code generation. 5
19
For this entry, the leaderboard records TianxiCode paired with DeepSeek-V4.1-Flash. Reporting describes the model as handling code understanding and reasoning, while TianxiCode provides the engineering workflow and tool use. The score therefore belongs to the combined system; it does not isolate the model’s contribution from the framework’s. 6
16
What “Verified” adds
QbitAI’s report says SWE-bench-Live’s Verified review requires full agent-run trajectories and checks evaluation inputs and information isolation for potential leakage of reference answers, tests, or results. TianxiCode’s entry passed that review. 5
That process offers more information about the evaluation than an unreviewed score alone. It should not be read as proof of broad real-world performance, however: the reported result remains tied to the Lite split and its benchmark tasks. 5
16
The framework’s reported role—and the limits of the evidence
Reports describe TianxiCode as supporting cross-file, multi-hop retrieval, autonomous planning with multiple tool calls, and closed-loop patch generation with test-driven correction. These are framework capabilities reported alongside the benchmark result, not separate measurements of how much each capability contributed to the 71% score. 9
15
The result is consistent with the idea that agent workflow and tooling can matter alongside the underlying model in coding benchmarks. It does not establish that a Flash-class model is more cost-efficient, quantify any savings, or show how TianxiCode would score with a different model. The available reporting also does not provide a specific integration timeline or named products for Lenovo’s broader deployment plans. 5
6
The takeaway
TianxiCode and DeepSeek-V4.1-Flash posted a strong, officially reviewed result on SWE-bench-Live Lite: 213 of 300 tasks resolved, with first place on the October 8 leaderboard. The clearest conclusion is also the narrowest one: this particular model-and-agent combination led the Lite board at that point, while broader benchmark leadership and real-world cost or deployment claims remain unproven by the score.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Lenovo’s TianxiCode Takes First on SWE-bench-Live Lite With a 71% Score"?
Lenovo’s TianxiCode coding agent framework, running DeepSeek V4.1 Flash, resolved 213 of 300 SWE bench Live Lite tasks for a verified 71% score and first place on the October 8, 2026 leaderboard.
What are the key points to validate first?
Lenovo’s TianxiCode coding agent framework, running DeepSeek V4.1 Flash, resolved 213 of 300 SWE bench Live Lite tasks for a verified 71% score and first place on the October 8, 2026 leaderboard. The benchmark tests agents on issues drawn from real GitHub projects; the score reflects the combined model and framework system, not DeepSeek alone.
What should I do next in practice?
Official verification included review of submitted agent trajectories and evaluation conditions intended to check for answer or test leakage.
Sources
forck.liveWeek 41, 2026 · AI lab announcements, 5–11 October 2026
swe-bench-live.github.ioSWE-bench-Live Leaderboard
qbitai.com联想天禧自研代码智能体TianxiCode斩获SWE-bench-Live全球第一
pandaily.comLenovo's TianxiCode Agent and DeepSeek-V4.1-Flash Top SWE-bench-Live Lite at 71%, Verified
x.comx.com › thePandaily › articleLenovo's TianxiCode Agent and DeepSeek-V4.1-Flash Top...