{"id":1044,"date":"2026-04-19T23:32:40","date_gmt":"2026-04-19T21:32:40","guid":{"rendered":"https:\/\/lewoniewski.info\/blog\/?p=1044"},"modified":"2026-05-31T16:24:08","modified_gmt":"2026-05-31T14:24:08","slug":"wyscig-modeli-ai-2024-2026","status":"publish","type":"post","link":"https:\/\/lewoniewski.info\/blog\/2026\/wyscig-modeli-ai-2024-2026\/","title":{"rendered":"Wy\u015bcig modeli AI 2024\u20132026 (wideo)"},"content":{"rendered":"<p>\u015awiat AI zmienia si\u0119 w zawrotnym tempie. W ci\u0105gu ostatnich dw\u00f3ch lat \u2014 od sierpnia 2024 do kwietnia 2026 \u2014 na podstawie aktualnych i historycznych danych rankingowych z platformy <a href=\"https:\/\/arena.ai\/\" target=\"_blank\" rel=\"noopener\">Arena AI<\/a> (wcze\u015bniej dzia\u0142aj\u0105cej pod nazwami LMArena i Chatbot Arena) \u015bledzono \u0142\u0105cznie <strong>347 du\u017cych modeli j\u0119zykowych<\/strong> (LLM\u00f3w). To jedna z najbardziej popularnych platform do spo\u0142eczno\u015bciowej oceny modeli AI.<\/p>\n<p>Arena AI stosuje nietypow\u0105 metod\u0119 oceny opart\u0105 na g\u0142osach u\u017cytkownik\u00f3w. Prawdziwi u\u017cytkownicy wpisuj\u0105 prompty i por\u00f3wnuj\u0105 obok siebie odpowiedzi dw\u00f3ch anonimowych modeli. Na podstawie tysi\u0119cy takich g\u0142os\u00f3w ka\u017cdy model otrzymuje wynik pokazuj\u0105cy jego wzgl\u0119dn\u0105 jako\u015b\u0107 w otwartych zadaniach tekst\u2013tekst. Im wy\u017cszy wynik, tym cz\u0119\u015bciej model wygrywa bezpo\u015brednie por\u00f3wnania z rywalami. W kwietniu 2026 skala wynik\u00f3w si\u0119ga od oko\u0142o 950 punkt\u00f3w (dla starszych modeli) do ponad 1500 (dla obecnej czo\u0142\u00f3wki).<\/p>\n<div class=\"see-also\">\n<p>Zobacz te\u017c: <a href=\"https:\/\/lewoniewski.info\/blog\/llm\/\">Ranking LLM: najlepsze modele AI<\/a><\/p>\n<\/div>\n<h3>Dynamiczna wizualizacja<\/h3>\n<p>Aby lepiej pokaza\u0107, jak zmienia\u0142a si\u0119 sytuacja, powsta\u0142a animacja w formie wy\u015bcigu na wykresie liniowym (tzw. <em>line race<\/em>), kt\u00f3ra przedstawia walk\u0119 czo\u0142owych modeli AI o najwy\u017csze pozycje w czasie. Gdy na scenie pojawiaj\u0105 si\u0119 nowe modele, a starsze stopniowo trac\u0105 znaczenie, kolejne przetasowania mo\u017cna obserwowa\u0107 niemal na \u017cywo \u2014 od wczesnej dominacji GPT-4o po wej\u015bcie na szczyt modeli Claude Opus 4.7 i Gemini 3 Pro. W animacji uwzgl\u0119dniono tak\u017ce specjalne kamienie milowe uruchamiane w momencie przekroczenia wa\u017cnych prog\u00f3w punktowych.<\/p>\n<p style=\"text-align:center;\">Ta dynamiczn\u0105 wizualizacja zosta\u0142a pokazana na filmiku na <a href=\"https:\/\/www.youtube.com\/watch?v=yySDnfkYH30\" rel=\"noopener\" target=\"_blank\">YouTube<\/a>: <iframe loading=\"lazy\" width=\"560\" height=\"315\" src=\"https:\/\/www.youtube.com\/embed\/yySDnfkYH30\" title=\"Odtwarzacz wideo YouTube\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/p>\n<h3>Modele, kt\u00f3re trafia\u0142y do pierwszej dziesi\u0105tki<\/h3>\n<p>Spo\u015br\u00f3d 347 ocenionych modeli tylko <strong>77 kiedykolwiek znalaz\u0142o si\u0119 w pierwszej dziesi\u0105tce<\/strong> rankingu og\u00f3lnego. Poni\u017csza tabela pokazuje ka\u017cdy z nich wraz z obecn\u0105 pozycj\u0105, najwy\u017cszym osi\u0105gni\u0119tym wynikiem i dat\u0105 ostatniej obecno\u015bci w \u015bcis\u0142ej czo\u0142\u00f3wce.<\/p>\n<table>\n<thead>\n<tr>\n<th style=\"width: 45%;\">Obecne miejsce \/ Model<\/th>\n<th>Obecny wynik<\/th>\n<th>Najwy\u017cszy wynik<\/th>\n<th style=\"width: 20%;\">Najlepsza pozycja w rankingu<\/th>\n<th>Ostatnio w Top 10<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>1. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-7\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7 Thinking<\/a><\/td>\n<td>1504,53<\/td>\n<td>1504,53<\/td>\n<td>1 <small>(2026-04-17)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>2. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-6\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.6 Thinking<\/a><\/td>\n<td>1502,63<\/td>\n<td>1506,96<\/td>\n<td>1 <small>(2026-04-14)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>3. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-7\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7<\/a><\/td>\n<td>1498,47<\/td>\n<td>1498,47<\/td>\n<td>3 <small>(2026-04-17)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>4. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-6\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.6<\/a><\/td>\n<td>1496,83<\/td>\n<td>1505,14<\/td>\n<td>1 <small>(2026-03-11)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>5. <a href=\"https:\/\/ai.meta.com\/blog\/introducing-muse-spark-msl\/\" target=\"_blank\" rel=\"noopener\">Muse Spark<\/a><\/td>\n<td>1495,88<\/td>\n<td>1495,88<\/td>\n<td>3 <small>(2026-04-14)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>6. <a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/gemini-3-1-pro\/\" target=\"_blank\" rel=\"noopener\">Gemini 3.1 Pro Preview<\/a><\/td>\n<td>1492,25<\/td>\n<td>1500,71<\/td>\n<td>2 <small>(2026-03-05)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>7. <a href=\"http:\/\/aistudio.google.com\/app\/prompts\/new_chat?model=gemini-3-pro-preview\" target=\"_blank\" rel=\"noopener\">Gemini 3 Pro<\/a><\/td>\n<td>1486,11<\/td>\n<td>1502,16<\/td>\n<td>1 <small>(2026-01-29)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>8. <a href=\"https:\/\/grok.com\" target=\"_blank\" rel=\"noopener\">Grok 4.20 Beta 1<\/a><\/td>\n<td>1485,01<\/td>\n<td>1496,02<\/td>\n<td>3 <small>(2026-03-11)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>9. <a href=\"https:\/\/platform.openai.com\/docs\/models\/gpt-5.4\" target=\"_blank\" rel=\"noopener\">GPT-5.4 High<\/a><\/td>\n<td>1481,63<\/td>\n<td>1485,70<\/td>\n<td>6 <small>(2026-04-07)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>10. <a href=\"https:\/\/docs.x.ai\/developers\/models\/grok-4.20-beta-0309-reasoning\" target=\"_blank\" rel=\"noopener\">Grok 4.20 Beta Reasoning (03-09)<\/a><\/td>\n<td>1479,81<\/td>\n<td>1483,48<\/td>\n<td>7 <small>(2026-04-07)<\/small><\/td>\n<td>2026-04-17<\/td>\n<\/tr>\n<tr>\n<td>11. <a href=\"https:\/\/developers.openai.com\/api\/docs\/models\/gpt-5.2-chat-latest\" target=\"_blank\" rel=\"noopener\">GPT-5.2 Chat Latest (2026-02-10)<\/a><\/td>\n<td>1477,12<\/td>\n<td>1502,50<\/td>\n<td>3 <small>(2026-02-17)<\/small><\/td>\n<td>2026-04-14<\/td>\n<\/tr>\n<tr>\n<td>12. <a href=\"https:\/\/docs.x.ai\/developers\/models\/grok-4.20-multi-agent-beta-0309\" target=\"_blank\" rel=\"noopener\">Grok 4.20 Multi-Agent Beta (03-09)<\/a><\/td>\n<td>1475,62<\/td>\n<td>1478,97<\/td>\n<td>9 <small>(2026-04-07)<\/small><\/td>\n<td>2026-04-14<\/td>\n<\/tr>\n<tr>\n<td>13. <a href=\"https:\/\/blog.google\/products\/gemini\/gemini-3-flash\" target=\"_blank\" rel=\"noopener\">Gemini 3 Flash<\/a><\/td>\n<td>1474,02<\/td>\n<td>1479,66<\/td>\n<td>2 <small>(2025-12-30)<\/small><\/td>\n<td>2026-04-07<\/td>\n<\/tr>\n<tr>\n<td>14. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-5\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.5 Thinking 32K (2025-11-01)<\/a><\/td>\n<td>1473,03<\/td>\n<td>1473,90<\/td>\n<td>3 <small>(2025-12-15)<\/small><\/td>\n<td>2026-03-31<\/td>\n<\/tr>\n<tr>\n<td>16. <a href=\"https:\/\/x.ai\/news\/grok-4-1\" target=\"_blank\" rel=\"noopener\">Grok 4.1 Thinking<\/a><\/td>\n<td>1469,85<\/td>\n<td>1484,41<\/td>\n<td>2 <small>(2026-01-29)<\/small><\/td>\n<td>2026-03-11<\/td>\n<\/tr>\n<tr>\n<td>17. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-5\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.5 (2025-11-01)<\/a><\/td>\n<td>1468,76<\/td>\n<td>1469,20<\/td>\n<td>3 <small>(2025-11-26)<\/small><\/td>\n<td>2026-03-06<\/td>\n<\/tr>\n<tr>\n<td>21. <a href=\"https:\/\/blog.google\/products\/gemini\/gemini-3-flash\" target=\"_blank\" rel=\"noopener\">Gemini 3 Flash (Thinking Minimal)<\/a><\/td>\n<td>1462,73<\/td>\n<td>1464,05<\/td>\n<td>7 <small>(2026-01-29)<\/small><\/td>\n<td>2026-02-11<\/td>\n<\/tr>\n<tr>\n<td>23. <a href=\"https:\/\/x.ai\/news\/grok-4-1\" target=\"_blank\" rel=\"noopener\">Grok 4.1<\/a><\/td>\n<td>1460,51<\/td>\n<td>1466,36<\/td>\n<td>3 <small>(2025-11-20)<\/small><\/td>\n<td>2026-02-11<\/td>\n<\/tr>\n<tr>\n<td>25. <a href=\"https:\/\/huggingface.co\/zai-org\/GLM-5\" target=\"_blank\" rel=\"noopener\">GLM-5<\/a><\/td>\n<td>1456,03<\/td>\n<td>1470,37<\/td>\n<td>8 <small>(2026-02-10)<\/small><\/td>\n<td>2026-02-10<\/td>\n<\/tr>\n<tr>\n<td>26. <a href=\"https:\/\/openai.com\/index\/gpt-5-1\/\" target=\"_blank\" rel=\"noopener\">GPT-5.1 High<\/a><\/td>\n<td>1454,71<\/td>\n<td>1460,58<\/td>\n<td>4 <small>(2025-11-20)<\/small><\/td>\n<td>2026-01-29<\/td>\n<\/tr>\n<tr>\n<td>27. <a href=\"https:\/\/openai.com\/index\/gpt-5-3-instant\/\" target=\"_blank\" rel=\"noopener\">GPT-5.3 Chat Latest<\/a><\/td>\n<td>1454,34<\/td>\n<td>1468,22<\/td>\n<td>10 <small>(2026-03-11)<\/small><\/td>\n<td>2026-03-11<\/td>\n<\/tr>\n<tr>\n<td>28. <a href=\"https:\/\/www.anthropic.com\/news\/claude-sonnet-4-5\" target=\"_blank\" rel=\"noopener\">Claude Sonnet 4.5 Thinking 32K (2025-09-29)<\/a><\/td>\n<td>1451,92<\/td>\n<td>1453,01<\/td>\n<td>1 <small>(2025-10-03)<\/small><\/td>\n<td>2026-01-29<\/td>\n<\/tr>\n<tr>\n<td>29. <a href=\"https:\/\/www.anthropic.com\/news\/claude-sonnet-4-5\" target=\"_blank\" rel=\"noopener\">Claude Sonnet 4.5 (2025-09-29)<\/a><\/td>\n<td>1451,66<\/td>\n<td>1452,74<\/td>\n<td>5 <small>(2025-11-09)<\/small><\/td>\n<td>2025-12-15<\/td>\n<\/tr>\n<tr>\n<td>32. <a href=\"https:\/\/ernie.baidu.com\/blog\/posts\/ernie-5.0-0110-release-on-lmarena\/\" target=\"_blank\" rel=\"noopener\">ERNIE 5.0 0110<\/a><\/td>\n<td>1450,44<\/td>\n<td>1453,72<\/td>\n<td>9 <small>(2026-01-29)<\/small><\/td>\n<td>2026-01-29<\/td>\n<\/tr>\n<tr>\n<td>33. <a href=\"https:\/\/ernie.baidu.com\/blog\/posts\/ernie-5.0-preview-1203-release-on-lmarena\/\" target=\"_blank\" rel=\"noopener\">ERNIE 5.0 Preview 1203<\/a><\/td>\n<td>1449,49<\/td>\n<td>1450,56<\/td>\n<td>9 <small>(2025-12-21)<\/small><\/td>\n<td>2025-12-21<\/td>\n<\/tr>\n<tr>\n<td>34. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-1\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.1 Thinking 16K (2025-08-05)<\/a><\/td>\n<td>1448,85<\/td>\n<td>1451,45<\/td>\n<td>2 <small>(2025-11-06)<\/small><\/td>\n<td>2026-01-09<\/td>\n<\/tr>\n<tr>\n<td>35. <a href=\"http:\/\/aistudio.google.com\/app\/prompts\/new_chat?model=gemini-2.5-pro\" target=\"_blank\" rel=\"noopener\">Gemini 2.5 Pro<\/a><\/td>\n<td>1448,66<\/td>\n<td>1466,64<\/td>\n<td>1 <small>(2025-11-09)<\/small><\/td>\n<td>2026-01-09<\/td>\n<\/tr>\n<tr>\n<td>36. <a href=\"https:\/\/www.anthropic.com\/news\/claude-opus-4-1\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.1 (2025-08-05)<\/a><\/td>\n<td>1446,83<\/td>\n<td>1462,10<\/td>\n<td>2 <small>(2025-08-07)<\/small><\/td>\n<td>2025-11-20<\/td>\n<\/tr>\n<tr>\n<td>39. <a href=\"https:\/\/openai.com\/index\/introducing-gpt-4-5\/\" target=\"_blank\" rel=\"noopener\">GPT-4.5 Preview (2025-02-27)<\/a><\/td>\n<td>1444,45<\/td>\n<td>1444,88<\/td>\n<td>1 <small>(2025-03-25)<\/small><\/td>\n<td>2025-11-20<\/td>\n<\/tr>\n<tr>\n<td>40. <a href=\"https:\/\/x.com\/OpenAI\/status\/1905331956856050135\" target=\"_blank\" rel=\"noopener\">ChatGPT-4o Latest (2025-03-26)<\/a><\/td>\n<td>1443,19<\/td>\n<td>1443,66<\/td>\n<td>1 <small>(2025-04-16)<\/small><\/td>\n<td>2025-11-09<\/td>\n<\/tr>\n<tr>\n<td>45. <a href=\"https:\/\/openai.com\/index\/gpt-5-1\/\" target=\"_blank\" rel=\"noopener\">GPT-5.1<\/a><\/td>\n<td>1438,68<\/td>\n<td>1440,92<\/td>\n<td>9 <small>(2025-11-16)<\/small><\/td>\n<td>2025-11-17<\/td>\n<\/tr>\n<tr>\n<td>47. <a href=\"https:\/\/www.alibabacloud.com\/help\/en\/model-studio\/models\" target=\"_blank\" rel=\"noopener\">Qwen3 Max Preview<\/a><\/td>\n<td>1434,94<\/td>\n<td>1435,12<\/td>\n<td>8 <small>(2025-09-30)<\/small><\/td>\n<td>2025-11-09<\/td>\n<\/tr>\n<tr>\n<td>49. <a href=\"https:\/\/platform.openai.com\/docs\/models\/gpt-5\" target=\"_blank\" rel=\"noopener\">GPT-5 High<\/a><\/td>\n<td>1433,37<\/td>\n<td>1481,37<\/td>\n<td>1 <small>(2025-08-18)<\/small><\/td>\n<td>2025-11-09<\/td>\n<\/tr>\n<tr>\n<td>52. <a href=\"https:\/\/openai.com\/index\/introducing-o3-and-o4-mini\/\" target=\"_blank\" rel=\"noopener\">o3 (2025-04-16)<\/a><\/td>\n<td>1431,27<\/td>\n<td>1454,32<\/td>\n<td>1 <small>(2025-06-18)<\/small><\/td>\n<td>2025-11-09<\/td>\n<\/tr>\n<tr>\n<td>55. <a href=\"https:\/\/platform.openai.com\/docs\/models\/gpt-5-chat-latest\" target=\"_blank\" rel=\"noopener\">GPT-5 Chat<\/a><\/td>\n<td>1426,56<\/td>\n<td>1429,60<\/td>\n<td>8 <small>(2025-09-08)<\/small><\/td>\n<td>2025-10-01<\/td>\n<\/tr>\n<tr>\n<td>60. <a href=\"https:\/\/www.anthropic.com\/news\/claude-4\" target=\"_blank\" rel=\"noopener\">Claude Opus 4 Thinking 16K (2025-05-14)<\/a><\/td>\n<td>1423,85<\/td>\n<td>1424,30<\/td>\n<td>6 <small>(2025-07-28)<\/small><\/td>\n<td>2025-09-18<\/td>\n<\/tr>\n<tr>\n<td>61. <a href=\"https:\/\/huggingface.co\/Qwen\/Qwen3-235B-A22B-Instruct-2507\" target=\"_blank\" rel=\"noopener\">Qwen3 235B-A22B Instruct 2507<\/a><\/td>\n<td>1423,50<\/td>\n<td>1432,93<\/td>\n<td>5 <small>(2025-08-04)<\/small><\/td>\n<td>2025-08-21<\/td>\n<\/tr>\n<tr>\n<td>64. <a href=\"https:\/\/api-docs.deepseek.com\/news\/news250528\" target=\"_blank\" rel=\"noopener\">DeepSeek R1 0528<\/a><\/td>\n<td>1421,98<\/td>\n<td>1421,98<\/td>\n<td>5 <small>(2025-06-18)<\/small><\/td>\n<td>2025-08-04<\/td>\n<\/tr>\n<tr>\n<td>65. <a href=\"https:\/\/x.ai\/news\/grok-4-fast\" target=\"_blank\" rel=\"noopener\">Grok 4 Fast Chat<\/a><\/td>\n<td>1421,08<\/td>\n<td>1424,78<\/td>\n<td>10 <small>(2025-09-30)<\/small><\/td>\n<td>2025-09-30<\/td>\n<\/tr>\n<tr>\n<td>70. <a href=\"https:\/\/moonshotai.github.io\/Kimi-K2\/\" target=\"_blank\" rel=\"noopener\">Kimi K2 Preview (07-11)<\/a><\/td>\n<td>1417,40<\/td>\n<td>1421,29<\/td>\n<td>6 <small>(2025-07-25)<\/small><\/td>\n<td>2025-08-28<\/td>\n<\/tr>\n<tr>\n<td>77. <a href=\"https:\/\/openai.com\/index\/gpt-4-1\/\" target=\"_blank\" rel=\"noopener\">GPT-4.1 (2025-04-14)<\/a><\/td>\n<td>1413,36<\/td>\n<td>1413,86<\/td>\n<td>4 <small>(2025-05-22)<\/small><\/td>\n<td>2025-07-15<\/td>\n<\/tr>\n<tr>\n<td>78. <a href=\"https:\/\/www.anthropic.com\/news\/claude-4\" target=\"_blank\" rel=\"noopener\">Claude Opus 4 (2025-05-14)<\/a><\/td>\n<td>1412,22<\/td>\n<td>1420,44<\/td>\n<td>4 <small>(2025-06-18)<\/small><\/td>\n<td>2025-08-01<\/td>\n<\/tr>\n<tr>\n<td>79. <a href=\"https:\/\/x.ai\/blog\/grok-3\" target=\"_blank\" rel=\"noopener\">Grok 3 Preview (02-24)<\/a><\/td>\n<td>1411,89<\/td>\n<td>1413,32<\/td>\n<td>2 <small>(2025-03-25)<\/small><\/td>\n<td>2025-07-28<\/td>\n<\/tr>\n<tr>\n<td>80. <a href=\"https:\/\/z.ai\/blog\/glm-4.5\" target=\"_blank\" rel=\"noopener\">GLM-4.5<\/a><\/td>\n<td>1411,16<\/td>\n<td>1418,42<\/td>\n<td>10 <small>(2025-08-04)<\/small><\/td>\n<td>2025-08-04<\/td>\n<\/tr>\n<tr>\n<td>81. <a href=\"http:\/\/aistudio.google.com\/app\/prompts\/new_chat?model=gemini-2.5-flash\" target=\"_blank\" rel=\"noopener\">Gemini 2.5 Flash<\/a><\/td>\n<td>1411,05<\/td>\n<td>1417,54<\/td>\n<td>6 <small>(2025-07-07)<\/small><\/td>\n<td>2025-07-17<\/td>\n<\/tr>\n<tr>\n<td>82. <a href=\"https:\/\/docs.x.ai\/docs\/models\/grok-4-0709\" target=\"_blank\" rel=\"noopener\">Grok 4 0709<\/a><\/td>\n<td>1410,12<\/td>\n<td>1436,78<\/td>\n<td>5 <small>(2025-07-28)<\/small><\/td>\n<td>2025-09-08<\/td>\n<\/tr>\n<tr>\n<td>89. <a href=\"https:\/\/qwenlm.github.io\/blog\/qwen3\/\" target=\"_blank\" rel=\"noopener\">Qwen3 235B-A22B No Thinking<\/a><\/td>\n<td>1403,21<\/td>\n<td>1403,21<\/td>\n<td>10 <small>(2025-07-07)<\/small><\/td>\n<td>2025-07-07<\/td>\n<\/tr>\n<tr>\n<td>93. <a href=\"https:\/\/openai.com\/index\/o1-and-new-tools-for-developers\/\" target=\"_blank\" rel=\"noopener\">o1 (2024-12-17)<\/a><\/td>\n<td>1401,79<\/td>\n<td>1402,44<\/td>\n<td>1 <small>(2025-02-27)<\/small><\/td>\n<td>2025-07-01<\/td>\n<\/tr>\n<tr>\n<td>98. <a href=\"https:\/\/api-docs.deepseek.com\/news\/news250120\" target=\"_blank\" rel=\"noopener\">DeepSeek R1<\/a><\/td>\n<td>1397,80<\/td>\n<td>1398,20<\/td>\n<td>2 <small>(2025-02-27)<\/small><\/td>\n<td>2025-05-22<\/td>\n<\/tr>\n<tr>\n<td>103. <a href=\"https:\/\/api-docs.deepseek.com\/news\/news250325\" target=\"_blank\" rel=\"noopener\">DeepSeek V3 0324<\/a><\/td>\n<td>1395,24<\/td>\n<td>1397,36<\/td>\n<td>4 <small>(2025-04-16)<\/small><\/td>\n<td>2025-06-18<\/td>\n<\/tr>\n<tr>\n<td>107. <a href=\"https:\/\/openai.com\/index\/introducing-o3-and-o4-mini\/\" target=\"_blank\" rel=\"noopener\">o4 Mini (2025-04-16)<\/a><\/td>\n<td>1389,90<\/td>\n<td>1400,07<\/td>\n<td>6 <small>(2025-05-11)<\/small><\/td>\n<td>2025-06-18<\/td>\n<\/tr>\n<tr>\n<td>109. <a href=\"https:\/\/www.anthropic.com\/news\/claude-4\" target=\"_blank\" rel=\"noopener\">Claude Sonnet 4 (2025-05-14)<\/a><\/td>\n<td>1388,86<\/td>\n<td>1395,42<\/td>\n<td>7 <small>(2025-06-11)<\/small><\/td>\n<td>2025-06-11<\/td>\n<\/tr>\n<tr>\n<td>110. <a href=\"https:\/\/platform.openai.com\/docs\/models\/o1\" target=\"_blank\" rel=\"noopener\">o1 Preview<\/a><\/td>\n<td>1387,97<\/td>\n<td>1388,54<\/td>\n<td>1 <small>(2024-12-22)<\/small><\/td>\n<td>2025-04-16<\/td>\n<\/tr>\n<tr>\n<td>114. <a href=\"https:\/\/www.anthropic.com\/news\/claude-3-7-sonnet\" target=\"_blank\" rel=\"noopener\">Claude 3.7 Sonnet Thinking 32K (2025-02-19)<\/a><\/td>\n<td>1386,76<\/td>\n<td>1388,93<\/td>\n<td>5 <small>(2025-03-26)<\/small><\/td>\n<td>2025-05-22<\/td>\n<\/tr>\n<tr>\n<td>125. <a href=\"https:\/\/qwenlm.github.io\/blog\/qwen2.5-max\/\" target=\"_blank\" rel=\"noopener\">Qwen2.5 Max<\/a><\/td>\n<td>1374,38<\/td>\n<td>1374,98<\/td>\n<td>5 <small>(2025-02-03)<\/small><\/td>\n<td>2025-03-25<\/td>\n<\/tr>\n<tr>\n<td>127. <a href=\"https:\/\/www.anthropic.com\/news\/3-5-models-and-computer-use\" target=\"_blank\" rel=\"noopener\">Claude 3.5 Sonnet (2024-10-22)<\/a><\/td>\n<td>1371,79<\/td>\n<td>1373,10<\/td>\n<td>2 <small>(2024-12-22)<\/small><\/td>\n<td>2025-03-26<\/td>\n<\/tr>\n<tr>\n<td>128. <a href=\"https:\/\/www.anthropic.com\/news\/claude-3-7-sonnet\" target=\"_blank\" rel=\"noopener\">Claude 3.7 Sonnet (2025-02-19)<\/a><\/td>\n<td>1370,74<\/td>\n<td>1375,92<\/td>\n<td>3 <small>(2025-02-27)<\/small><\/td>\n<td>2025-04-16<\/td>\n<\/tr>\n<tr>\n<td>134. <a href=\"https:\/\/platform.openai.com\/docs\/guides\/reasoning#reasoning-effort\" target=\"_blank\" rel=\"noopener\">o3 Mini High<\/a><\/td>\n<td>1363,44<\/td>\n<td>1365,82<\/td>\n<td>4 <small>(2025-02-21)<\/small><\/td>\n<td>2025-04-16<\/td>\n<\/tr>\n<tr>\n<td>137. <a href=\"https:\/\/aistudio.google.com\/app\/prompts\/new_chat?instructions=lmsys-1121&amp;model=gemini-2.0-flash-001\" target=\"_blank\" rel=\"noopener\">Gemini 2.0 Flash 001<\/a><\/td>\n<td>1360,16<\/td>\n<td>1366,27<\/td>\n<td>4 <small>(2025-02-06)<\/small><\/td>\n<td>2025-03-17<\/td>\n<\/tr>\n<tr>\n<td>138. <a href=\"https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3\" target=\"_blank\" rel=\"noopener\">DeepSeek V3<\/a><\/td>\n<td>1358,40<\/td>\n<td>1358,99<\/td>\n<td>4 <small>(2025-01-22)<\/small><\/td>\n<td>2025-02-27<\/td>\n<\/tr>\n<tr>\n<td>145. <a href=\"https:\/\/aistudio.google.com\/prompts\/new_chat?model=gemini-2.0-flash-lite\" target=\"_blank\" rel=\"noopener\">Gemini 2.0 Flash Lite Preview (02-05)<\/a><\/td>\n<td>1353,09<\/td>\n<td>1353,65<\/td>\n<td>10 <small>(2025-02-17)<\/small><\/td>\n<td>2025-02-17<\/td>\n<\/tr>\n<tr>\n<td>146. <a href=\"https:\/\/aistudio.google.com\/app\/prompts\/new_chat?model=gemini-1.5-pro-002\" target=\"_blank\" rel=\"noopener\">Gemini 1.5 Pro 002<\/a><\/td>\n<td>1350,91<\/td>\n<td>1351,71<\/td>\n<td>2 <small>(2024-10-23)<\/small><\/td>\n<td>2025-02-27<\/td>\n<\/tr>\n<tr>\n<td>150. <a href=\"https:\/\/openai.com\/index\/openai-o3-mini\/\" target=\"_blank\" rel=\"noopener\">o3 Mini<\/a><\/td>\n<td>1347,62<\/td>\n<td>1348,61<\/td>\n<td>8 <small>(2025-02-14)<\/small><\/td>\n<td>2025-02-21<\/td>\n<\/tr>\n<tr>\n<td>158. <a href=\"https:\/\/openai.com\/index\/hello-gpt-4o\/\" target=\"_blank\" rel=\"noopener\">GPT-4o (2024-05-13)<\/a><\/td>\n<td>1345,41<\/td>\n<td>1346,08<\/td>\n<td>1 <small>(2024-09-15)<\/small><\/td>\n<td>2025-02-06<\/td>\n<\/tr>\n<tr>\n<td>161. <a href=\"https:\/\/www.anthropic.com\/news\/claude-3-5-sonnet\" target=\"_blank\" rel=\"noopener\">Claude 3.5 Sonnet (2024-06-20)<\/a><\/td>\n<td>1341,69<\/td>\n<td>1343,37<\/td>\n<td>2 <small>(2024-09-15)<\/small><\/td>\n<td>2025-02-05<\/td>\n<\/tr>\n<tr>\n<td>165. <a href=\"https:\/\/platform.openai.com\/docs\/models\/o1\" target=\"_blank\" rel=\"noopener\">o1 Mini<\/a><\/td>\n<td>1336,86<\/td>\n<td>1337,32<\/td>\n<td>2 <small>(2024-09-27)<\/small><\/td>\n<td>2025-02-03<\/td>\n<\/tr>\n<tr>\n<td>168. <a href=\"https:\/\/x.ai\/blog\/grok-2\" target=\"_blank\" rel=\"noopener\">Grok 2 (2024-08-13)<\/a><\/td>\n<td>1335,11<\/td>\n<td>1335,58<\/td>\n<td>5 <small>(2024-08-28)<\/small><\/td>\n<td>2024-12-05<\/td>\n<\/tr>\n<tr>\n<td>169. <a href=\"https:\/\/platform.openai.com\/docs\/models\/gpt-4o\" target=\"_blank\" rel=\"noopener\">GPT-4o (2024-08-06)<\/a><\/td>\n<td>1334,65<\/td>\n<td>1335,39<\/td>\n<td>7 <small>(2024-09-15)<\/small><\/td>\n<td>2024-12-22<\/td>\n<\/tr>\n<tr>\n<td>170. <a href=\"https:\/\/gemini.google.com\/advanced\" target=\"_blank\" rel=\"noopener\">Gemini Advanced 0514<\/a><\/td>\n<td>1334,59<\/td>\n<td>1335,24<\/td>\n<td>3 <small>(2024-09-15)<\/small><\/td>\n<td>2025-01-28<\/td>\n<\/tr>\n<tr>\n<td>171. <a href=\"https:\/\/ai.meta.com\/blog\/meta-llama-3-1\/\" target=\"_blank\" rel=\"noopener\">Llama 3.1 405B Instruct Bf16<\/a><\/td>\n<td>1334,54<\/td>\n<td>1335,95<\/td>\n<td>4 <small>(2024-09-15)<\/small><\/td>\n<td>2025-01-24<\/td>\n<\/tr>\n<tr>\n<td>173. <a href=\"https:\/\/ai.meta.com\/blog\/meta-llama-3-1\/\" target=\"_blank\" rel=\"noopener\">Llama 3.1 405B Instruct Fp8<\/a><\/td>\n<td>1332,73<\/td>\n<td>1334,26<\/td>\n<td>5 <small>(2024-09-15)<\/small><\/td>\n<td>2025-01-05<\/td>\n<\/tr>\n<tr>\n<td>181. <a href=\"https:\/\/platform.openai.com\/docs\/models\/gpt-4-turbo-and-gpt-4\" target=\"_blank\" rel=\"noopener\">GPT-4 Turbo (2024-04-09)<\/a><\/td>\n<td>1323,73<\/td>\n<td>1324,88<\/td>\n<td>8 <small>(2024-09-15)<\/small><\/td>\n<td>2024-09-27<\/td>\n<\/tr>\n<tr>\n<td>187. <a href=\"https:\/\/www.anthropic.com\/news\/claude-3-family\" target=\"_blank\" rel=\"noopener\">Claude 3 Opus (2024-02-29)<\/a><\/td>\n<td>1321,03<\/td>\n<td>1323,08<\/td>\n<td>9 <small>(2024-09-15)<\/small><\/td>\n<td>2024-09-15<\/td>\n<\/tr>\n<tr>\n<td>&#8211; ChatGPT-4o Latest<\/td>\n<td>&#8211;<\/td>\n<td>1288,84<\/td>\n<td>1 <small>(2024-09-04)<\/small><\/td>\n<td>2024-09-04<\/td>\n<\/tr>\n<tr>\n<td>&#8211; Dola Seed 2.0 Preview<\/td>\n<td>&#8211;<\/td>\n<td>1474,55<\/td>\n<td>7 <small>(2026-02-24)<\/small><\/td>\n<td>2026-03-04<\/td>\n<\/tr>\n<tr>\n<td>&#8211; Gemini 1.5 Pro API 0514<\/td>\n<td>&#8211;<\/td>\n<td>1238,75<\/td>\n<td>10 <small>(2024-09-15)<\/small><\/td>\n<td>2024-09-15<\/td>\n<\/tr>\n<tr>\n<td>&#8211; Llama 3.1 405B Instruct<\/td>\n<td>&#8211;<\/td>\n<td>1250,04<\/td>\n<td>5 <small>(2024-09-04)<\/small><\/td>\n<td>2024-09-04<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Kilka ciekawych obserwacji:<\/p>\n<ul>\n<li><strong>Anthropic dominuje w obecnym top 3.<\/strong> Claude Opus 4.7 Thinking zajmuje pierwsze miejsce z wynikiem 1504.53 \u2014 to pierwszy model w historii, kt\u00f3ry przekroczy\u0142 granic\u0119 1500 punkt\u00f3w na Arena AI. Wersja bez trybu \u201eThinking\u201d, czyli Claude Opus 4.7, jest trzecia, a Claude Opus 4.6 Thinking plasuje si\u0119 na miejscu drugim.<\/li>\n<li><strong>Google i xAI s\u0105 tu\u017c za liderami.<\/strong> Gemini 3 Pro, Gemini 3.1 Pro Preview i Gemini 3 Flash znajduj\u0105 si\u0119 obecnie w top 15. Z kolei Grok 4.20 Beta 1 (#5) oraz kilka innych wariant\u00f3w Groka pokazuj\u0105, jak wyr\u00f3wnana jest ta rywalizacja.<\/li>\n<li><strong>Historia flagowych modeli OpenAI jest wyj\u0105tkowo dynamiczna.<\/strong> GPT-5 High by\u0142 kiedy\u015b numerem 1 \u2014 w sierpniu 2025 osi\u0105gn\u0105\u0142 wynik 1481 punkt\u00f3w \u2014 ale od tamtej pory spad\u0142 na 49. miejsce, gdy wyprzedzi\u0142y go nowsze modele. Najnowszy model OpenAI, GPT-5.4 High, zajmuje obecnie 6. pozycj\u0119.<\/li>\n<li><strong>\u201eByli mistrzowie\u201d pokazuj\u0105 skal\u0119 zmian.<\/strong> GPT-4o by\u0142 numerem 1 jeszcze we wrze\u015bniu 2024, a dzi\u015b znajduje si\u0119 dopiero na 158. miejscu. o1 Preview prowadzi\u0142 w grudniu 2024, a teraz jest na 110. pozycji.<\/li>\n<li><strong>Chi\u0144skie laboratoria AI coraz mocniej zaznaczaj\u0105 swoj\u0105 obecno\u015b\u0107.<\/strong> GLM-5.1, Qwen3.5 Max Preview, ERNIE 5.0 i Kimi K2.5 Thinking pojawiaj\u0105 si\u0119 w top 20, co pokazuje, \u017ce Zhipu AI, Alibaba, Baidu i Moonshot s\u0105 dzi\u015b pe\u0142noprawnymi uczestnikami tego wy\u015bcigu.<\/li>\n<li><strong>Modele open source tak\u017ce dociera\u0142y do \u015bcis\u0142ej czo\u0142\u00f3wki.<\/strong> DeepSeek R1 osi\u0105gn\u0105\u0142 szczyt na 2. miejscu w lutym 2025, a DeepSeek V3.2 Exp Thinking znajduje si\u0119 obecnie na 10. pozycji \u2014 to bardzo mocne wyniki jak na modele z otwart\u0105 licencj\u0105.<\/li>\n<li><strong>G\u00f3rna granica wynik\u00f3w stale ro\u015bnie.<\/strong> W sierpniu 2024 lider mia\u0142 oko\u0142o 1290 punkt\u00f3w. W kwietniu 2026 najlepszy model przekracza ju\u017c 1504 punkty. To wzrost o oko\u0142o 17% w mniej ni\u017c dwa lata, odzwierciedlaj\u0105cy szybki post\u0119p w architekturze modeli, jako\u015bci danych treningowych i metodach dostrajania.<\/li>\n<li><strong>Czas utrzymywania si\u0119 w top 10 wyra\u017anie si\u0119 skraca.<\/strong> Wcze\u015bniejsze modele, takie jak GPT-4o czy Gemini 1.5 Pro, potrafi\u0142y utrzymywa\u0107 si\u0119 w pierwszej dziesi\u0105tce przez wiele miesi\u0119cy. Nowsze modele bywaj\u0105 wypierane ju\u017c po kilku tygodniach, bo tempo premier znacz\u0105co wzros\u0142o.<\/li>\n<\/ul>\n<h3>Metodologia<\/h3>\n<p>Ta analiza opiera si\u0119 na ocenach z kategorii \u201eoverall\u201d platformy Arena AI dla zada\u0144 typu tekst\u2013tekst. Pozycje w rankingu opieraj\u0105 si\u0119 na wynikach uzyskanych z ocen par modeli dokonywanych przez u\u017cytkownik\u00f3w. Przeanalizowano dane z okresu od 28 sierpnia 2024 do 17 kwietnia 2026. Do przedstawionej wcze\u015bniej tabeli (patrz sekcj\u0119 &#8222;Modele, kt\u00f3re trafia\u0142y do pierwszej dziesi\u0105tki&#8221;) zosta\u0142y wybrane modele, kt\u00f3re cho\u0107 raz znalaz\u0142y si\u0119 w pierwszej dziesi\u0105tce w tym przedziale czasu. Warto\u015bci okre\u015blone jako \u201eobecny wynik\u201d odnosz\u0105 si\u0119 do najnowszych dost\u0119pnych danych.<\/p>\n<p>\u0179r\u00f3d\u0142o danych: <a href=\"https:\/\/arena.ai\/leaderboard\/text\" target=\"_blank\" rel=\"noopener\">Arena AI (Text Arena)<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u015awiat AI zmienia si\u0119 w zawrotnym tempie. W ci\u0105gu ostatnich dw\u00f3ch lat \u2014 od sierpnia 2024 do kwietnia 2026 \u2014 na podstawie aktualnych i historycznych danych rankingowych z platformy Arena AI (wcze\u015bniej dzia\u0142aj\u0105cej pod nazwami LMArena i Chatbot Arena) \u015bledzono \u0142\u0105cznie 347 du\u017cych modeli j\u0119zykowych (LLM\u00f3w). To jedna z najbardziej popularnych platform do spo\u0142eczno\u015bciowej oceny [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":1046,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[12],"tags":[128,199,200,197,202,196,203,206,201,130,8,187],"class_list":["post-1044","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artykuly","tag-chatgpt","tag-claude","tag-deepseek","tag-grok","tag-llama","tag-llm","tag-mistral","tag-muse-spark","tag-qwen","tag-sztuczna-inteligencja","tag-wideo","tag-wizualizacja-danych"],"_links":{"self":[{"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/posts\/1044","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/comments?post=1044"}],"version-history":[{"count":0,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/posts\/1044\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/media\/1046"}],"wp:attachment":[{"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/media?parent=1044"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/categories?post=1044"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lewoniewski.info\/blog\/wp-json\/wp\/v2\/tags?post=1044"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}