IFM/K2-Horizon-7B

K2-Horizon-7B is the medium dense member of the K2-Horizon family: a 7B-core decoder-only model with a 512K context window. The chart at the top of this card shows K2-Horizon-7B against selected reference models. The table below lists every comparison model used in the figure. Reference models · weak to strongBenchmarkK2-Horizon-7BReference 1Reference 2Reference 3MathHMMT Feb 2026Competition mathematics73.3Gemma 4-12B63.1Qwen3.5-9B65.7Granite 4.2-8B66.5CodingSWE-bench VerifiedSoftware engineering70.6Gemma 4-12B30.6Granite 4.2-8B47.7Qwen3.5-9B50.8Scientific ReasoningHLEExpert-level reasoning18.6Granite 4.2-8B9.7Qwen3.5-9B14.9Gemma 4-12B15.7CodingSciCodeScientific coding31.6Qwen3.5-9B27.5Mistral Small 428.0Granite 4.2-8B30.4GeneralLCRLong-context reasoning68.0Granite 4.2-8B43.3Gemma 4-12B61.7Qwen3.5-9B65.3CodingTerminal-Bench 2.1Agentic terminal use39.1Granite 4.2-8B18.4Gemma 4-12B27.3Qwen3.5-9B29.2Agentstau3-BankingAgentic tool use25.8Qwen3.5-9B7.0Granite 4.2-8B7.6Muse Glimmer-30B24.0BrowseCompWeb browsing59.0DeepSeek V4 Flash-042353.5GPT-554.9LongCat Flash Thinking-260156.6 Scores in %. Bold marks the best score in each row. BrowseComp: our model uses the Discard-all@95k context-length protocol proposed in the DeepSeek-V3.2 technical report; comparison models may use different harnesses. Small reasoning models occasionally fall into degenerate repetition ("doom loops") on long generations and keep generating until they hit the output limit. After post-training, the 3.7B and 7B models receive a short…