Xiaomi updates MiMo-V2-Flash for coding agents
Xiaomi's January MiMo-V2-Flash update adds Thinking-mode support in Claude Code and improves Arena-Hard scores, while SWE-Bench Verified results stay roughly unchanged at about 73–74.
The page is titled as an announcement of three new MiMo-V2.6 models. However, the only content Xiaomi published there is an older release note, dated January 12, 2026, for an update to MiMo-V2-Flash. The page gives no details on any V2.6 models, including names, specs or benchmarks.
The note describes mimo-v2-flash-0112 as an incremental update. Xiaomi says it improves general capabilities and code generation in Thinking mode. It also says the model now fully supports Thinking mode in Claude Code and recommends making Thinking the default there for "more stable, higher-quality code generation." The company reports matching improvements in other coding-agent tools, including Kilo, Cline and Roo, and says the model is more stable and follows output-format instructions more closely.
Xiaomi's figures show modest gains:
| Benchmark | 0112 | Previous | |---|---|---| | SWE-Bench Verified (Non-Thinking) | 73.3 | 73.4 | | SWE-Bench Verified (Thinking) | 74.2 | — | | Arena-Hard Hard Prompt (Non-Thinking) | 52.7 | 46.0 | | Arena-Hard Creative Writing (Non-Thinking) | 86.0 | 78.3 | | Arena-Hard Hard Prompt (Thinking) | 58.3 | 54.1 | | Arena-Hard Creative Writing (Thinking) | 90.4 | 86.2 |
Coding performance without Thinking mode is flat. Most of the gains are on Arena-Hard. All figures come from Xiaomi and have not been independently verified.
Xiaomi says the API call method and model name are unchanged, so existing integrations should not need changes.