Gemini 2.5 Pro Benchmark Table

Gemini 2.5 Pro Preview 05-06 Thinking mode ON OpenAI o3 OpenAI o4-mini Claude Opus 4 Grok 3 Beta DeepSeek R1 Benchmark Input price Output price Reasoning & knowledge Humanity’s Last Exam (no tools) Science GPQA diamond Mathematics AIME 2025 Code generation LiveCodeBench Code editing Aider Polyglot Agentic coding SWE-bench Verified Factuality SimpleQA Factuality FACTS Grounding Visual reasoning MMMU Image understanding Ego-Wan (Reka) Video understanding VideoMMMU Long context MRCR v2 (1–needle) M.ultilingual understanding Global MMLU (Lite) $1.25 $10.00 $10.00 $40.00 $1.10 $4.40 $15.00 $75.00 $3.00 $15.00 $0.55 $2.19 21.6% 20.3% 14.3% 10.7% — 14.0%* 86.4% 83.3% 81.4% 79.6% 80.2% 81.0% 88.0% 88.9% 92.7% 75.5% 77.3% 87.5% 90.0% 93.3% 69.0% 72.0% 75.8% 51.1% — 70.5% 82.2% 79.6% 72.0% 72.0% 53.3% 71.6% 59.6% 49.4% 68.1% 72.5% — 57.6% 67.2% 79.4% 54.0% 48.6% 19.3% — 43.6% 27.8% 87.8% 69.6% 62.1% 77.7% 74.8% — 82.0% 82.9% 81.6% 76.5% 76.0% no VLM support 67.2% — — — — no VLM support 83.6% — — — — no VLM support 58.0% 57.1% 36.3% — 34.0% — 16.4% no support no support no support no support — — 89.2% — — — — — The results shown in this table are based on the latest available data. Multiple attempts settings allow test-time compute; results may vary. * Gemini 2.5 Pro: The scores are based on the latest available version of the model, with thinking mode enabled. ** Pricing is based on input and output tokens and may vary by region and usage tier.

A screenshot shows a benchmark comparison table for Gemini 2.5 Pro, OpenAI models, Claude Opus 4, Grok 3 Beta, and DeepSeek R1. The joke is about Google naming its new flagship model “05-06,” which looks like an awkward date or version number despite the model’s benchmark claims. It is shared as a sarcastic reaction to confusing AI model naming and release announcements.

• @dev_meme / 6853 2025-06-05