V4-Pro ranks first on Codeforces 3206 by beating GPT-5.4, but with long context and knowledge, it still falls behind Opus and Gemini

According to Beating Monitoring, the V4 technical report has released a comparison between DeepSeek-V4-Pro-Max (the highest inference strength mode) and a closed-source flagship. The comparison set includes Opus 4.6 Max, GPT-5.4 xHigh, Gemini 3.1 Pro High, as well as the open-source models Kimi K2.6 and GLM-5.1, excluding the recently released Opus 4.7 and GPT-5.5.

In terms of coding, V4-Pro-Max scored 3206 on Codeforces, exceeding GPT-5.4’s 3168 and Gemini 3.1 Pro’s 3052, setting a new record for this benchmark. On LiveCodeBench, the score of 93.5 was also the highest across the board. SWE Verified scored 80.6, only 0.2 percentage points lower than Opus 4.6’s 80.8.

For long-context benchmarks, both 1M tests ranked V4-Pro-Max second: CorpusQA 1M scored 62.0, trailing Opus 4.6’s 71.7 but ahead of Gemini 3.1 Pro’s 53.8; MRCR 1M scored 83.5, with Opus 4.6 leading at 92.9 by nearly 10 percentage points.

For agent tasks, MCPAtlas Public scored 73.6, just below Opus 4.6’s 73.8. Terminal-Bench 2.0 scored 67.9, lower than GPT-5.4’s 75.1 and Gemini 3.1 Pro’s 68.5.

In knowledge and reasoning, V4-Pro-Max still shows a clear gap: GPQA Diamond scored 90.1 (Gemini 94.3), SimpleQA-Verified scored 57.9 (Gemini 75.6), and HLE scored 37.7 (Gemini 44.4). As an open-source model, V4-Pro-Max is the first to match and even surpass the closed-source flagship on multiple coding and long-context benchmarks; however, it still lags behind Gemini 3.1 Pro on knowledge-intensive evaluations.

It should be noted that the comparison above does not include the recently released GPT-5.5 and Opus 4.7, and the performance gap between V4 and the latest generation of closed-source models needs to be verified by third-party testing.