ÖNCÜ LLM KARŞILAŞTIRMA · 1 EKİM 2026

Modeller Arası Başarım: Claude · GPT · Gemini · Kimi · GLM · Qwen · DeepSeek · MiniMax · Grok · Fable

Her satır kendi içinde renklendirildi: yeşil = yüksek, sarı = orta, kırmızı = düşük. Birinci en yeşil + beyaz çerçeveli. Boş hücreler veri yayınlanmadığını gösterir.
düşük yüksek · birinci beyaz çerçeveli · gri kesikli çerçeve = Mythos Preview (yayınlanmamış referans, birinciliğe dahil değil)
KAZANÇ SAYISI — En yüksek skoru aldığı benchmark sayısı (Mythos Preview referans, hariç)
20
Fable 5
9
Opus 5
6
Gemini 3.1 Pro
4
Opus 4.8
4
Opus 4.6
4
GPT-5.4
3
Fable 5.1
3
GPT-5.6 Sol
3
DeepSeek V4-Pro
3
Gemini 3.5 Flash
ANTHROPIC OPENAI GOOGLE MOONSHOT Z.AI ALIBABA DEEPSEEK MINIMAX xAI META
Opus 5.5(Eyl 26) Fable
5.1(Eyl 26)
Fable
5(Haz 26)
Mythos
Preview(Nis 26)
Opus
5(Tem 26)
Opus
4.8(May 26)
Opus
4.7(Nis 26)
Opus
4.6(Şub 26)
Opus
4.5(Kas 25)
Sonnet 5.5(Eyl 26) Sonnet
5(Haz 26)
Sonnet
4.6(Şub 26)
GPT 6.1 Sol(Eyl 26) GPT
6 Astra(Eyl 26)
GPT
5.6 Sol(Haz 26)
GPT
5.5(Nis 26)
GPT
5.4(Mar 26)
Gemini 4
Argon(Eyl 26)
Gemini
3.8 Flash(Eyl 26)
Gemini
3.7 Flash(Ağu 26)
Gemini
3.6 Flash(Tem 26)
Gemini
3.5 Flash(May 26)
Gemini
3.1 Pro(Şub 26)
Gemini
3 Pro(Kas 25)
Kimi
K3(Tem 26)
Kimi
K2.6(Nis 26)
GLM
5.3(Ağu 26)
GLM
5.2(Haz 26)
GLM
5.1(Mar 26)
Qwen
3.8 Max(Ağu 26)
Qwen
3.7 Max(May 26)
Qwen
3.6 Plus(Nis 26)
V4.1
Flash(Eyl 26)
V4-Pro
0813(Ağu 26)
V4-Pro(Nis 26) V3.2(Ara 25) V3(Ara 24) M3(Haz 26) Grok 4.7(Eyl 26) Grok
4.6(Ağu 26)
Grok
4.5(Tem 26)
Grok
4.3(Nis 26)
Grok
4.20(Şub 26)
Grok
4(Tem 25)
Muse Spark
1.3(Eyl 26)
Muse Spark
1.2(Ağu 26)
Muse Spark
1.1(Tem 26)
YAZILIM GELİŞTİRME / KODLAMA
SWE-bench Verified——95.5%93.9%96.0%88.6%87.6%80.8%80.9%—85.2%79.6%————77.2%—————80.6%76.2%———————78.8%——80.6%—42.0%80.5%—————72.0%———
SWE-bench Pro—81.2%80.3%77.8%79.2%69.2%64.3%53.4%——63.2%————58.6%57.7%————55.1%54.2%43.3%———62.1%———56.6%——55.4%——59.0%——64.7%—————61.5%
SWE-bench Multilingual—89.1%86.6%87.3%89.5%84.4%80.5%77.5%——78.3%————————————————————73.8%——76.2%—————78.0%——————
SWE-bench Multimodal—54.7%54.1%59.0%59.4%38.4%34.5%27.1%———————————————————————————————————————
DeepSWE v1.1——70.0%—68.8%59.0%————54.0%30.0%—74.1%73.0%67.0%52.0%—73.7%65.3%49.0%37.0%12.0%—69.0%—66.9%44.0%—56.6%—————————65.9%54.0%———75.4%59.3%53.0%
Terminal-Bench 2.0———82.0%——69.4%65.4%59.3%——59.1%———82.7%75.1%—————68.5%56.9%———————61.6%——67.9%46.4%———————————
Terminal-Bench 2.1——88.0%78.0%89.1%82.7%66.1%———80.4%———88.8%83.4%———85.8%—76.2%70.7%—88.3%—88.2%—————90.6%87.9%———66.0%————————80.0%
Terminal-Bench 4.0—57.9%44.5%—51.8%23.6%————12.4%——58.2%37.3%———19.1%11.2%——————41.8%————————————20.3%12.4%——————
Terminal-Bench Hard——62.9%——58.3%51.5%46.2%47.0%——53.0%——65.9%60.6%57.6%————39.4%53.8%41.7%—43.9%—50.8%43.2%—50.8%43.9%——46.2%35.6%6.8%42.4%———37.9%37.9%37.9%———
Terminal-Bench-Science 0.1——21.4%—30.0%10.5%————————22.4%—————————7.1%—8.1%————————————7.1%———————
FrontierCode (Diamond)——29.3%——13.4%—————————5.7%———————————————————————————————
CursorBench 3.2—73.4%70.5%—70.0%62.3%————61.5%———67.2%58.4%——69.2%61.6%————60.8%——55.0%———————————70.8%———————
LiveCodeBench———————88.8%——————————————91.7%————————87.1%——93.5%49.2%40.5%——————79.0%———
Codeforces (Elo)————————————————3168—————3052———————————3206————————————
GENEL BİLGİ & AKIL YÜRÜTME
▸ Fen & Bilim
GPQA Diamond—93.7%92.6%—93.2%92.0%91.4%89.6%86.6%—91.1%87.5%—96.1%94.1%93.5%92.0%—95.3%92.1%92.8%92.1%94.1%90.8%93.5%91.1%91.7%89.5%86.8%92.8%92.3%88.2%—92.8%88.8%84.0%55.7%92.9%—94.9%93.1%90.1%91.1%87.7%93.5%90.4%89.8%
CritPt31.7%29.7%28.6%—29.1%20.9%12.0%12.6%4.6%31.4%16.9%3.1%31.7%31.7%32.3%27.1%23.4%27.1%18.3%9.4%10.6%10.9%17.7%9.1%23.4%8.0%19.1%20.9%4.6%17.7%13.4%2.9%14.3%18.0%12.9%2.9%0.0%3.7%17.7%17.1%15.4%8.0%6.6%2.0%24.9%17.7%15.1%
SciCode66.9%63.1%61.0%—56.4%54.4%54.5%51.9%49.5%61.0%54.3%50.1%54.2%56.5%57.1%55.8%56.6%61.8%56.6%59.8%53.4%53.1%58.9%56.1%59.5%51.5%59.0%51.2%44.8%52.1%49.5%40.7%51.9%51.0%50.0%38.9%35.4%47.1%57.4%56.5%55.0%48.3%45.6%45.7%58.8%57.4%58.8%
Humanity's Last Exam61.4%59.1%55.5%—54.9%48.7%42.3%39.9%30.1%55.0%41.3%33.6%52.9%54.7%49.5%45.8%43.7%57.1%47.8%39.0%40.8%41.3%44.7%37.2%46.9%37.5%42.3%41.1%30.1%43.1%40.5%27.8%39.2%41.0%35.9%22.2%3.6%39.0%43.1%42.9%42.7%37.2%32.2%23.9%48.7%45.5%46.2%
Blueprint-Bench 2——38.6%——14.5%24.5%————6.7%———36.2%—————33.6%26.5%————————————————————————
▸ Sağlık & Biyomedikal
HealthBench——62.7%61.1%57.8%59.3%————58.7%———57.0%56.5%———————————————————————————————
HealthBench Professional——66.0%64.7%59.8%56.9%————57.8%———60.5%51.8%———————————————————————————————
BioMysteryBench (insan)——83.9%82.6%—80.4%—————————————————————————————————————————
BioMysteryBench (zor)——46.1%29.6%—40.0%—————————————————————————————————————————
▸ Bilgi & Gerçeklik
MMLU-Pro———————89.1%————————87.5%—————91.0%————————88.5%——87.5%81.2%75.9%——————————
MMMLU (Çok Dilli)———92.7%——91.5%91.1%90.8%——89.3%——————————92.6%91.8%———————89.5%———————————————
SimpleQA-Verified———————46.2%————————45.3%—————75.6%———————————57.9%—24.9%——————————
Chinese-SimpleQA———————76.2%————————76.8%—————85.9%———————————84.4%—64.8%——————————
IFBench——63.5%——62.2%58.6%53.1%58.0%——56.6%——72.7%75.9%73.9%————74.6%77.1%70.4%—76.0%—73.3%76.3%—80.5%75.2%——76.5%60.7%34.8%82.9%———81.3%81.2%53.7%———
▸ Soyut Akıl Yürütme (ARC-AGI)
ARC-AGI-1—97.5%98.5%91.2%97.5%92.5%93.5%93.0%80.0%—87.5%86.5%——97.5%95.0%93.7%——87.2%—92.5%98.0%75.0%94.5%——77.0%———————57.0%——————89.5%66.6%———
ARC-AGI-2—90.0%89.2%60.4%90.4%72.1%75.8%68.8%37.6%—67.1%60.4%——92.5%85.0%74.0%——52.6%—72.1%77.1%31.1%60.4%——22.8%———————4.0%——————65.1%15.9%———
ARC-AGI-3——2.1%0.3%30.2%1.5%0.2%0.5%——————7.8%0.4%0.2%—————0.4%———————————————————0.1%————
▸ Çok Biçimli & Görsel
MMMU-Pro (Çok Modal)———————73.9%———74.5%———81.2%81.2%————83.6%80.5%81.0%—————————————78.1%—————————
CharXiv Reasoning (araçsız)——88.9%86.1%——82.1%69.1%—————————————84.2%—————————————————————————
CharXiv Reasoning (araçlı)——93.5%93.2%——91.0%84.7%———————————————————————————————————————
ChartQAPro (araçsız)—————69.4%67.6%————————————————————————————————————————
ChartQAPro (araçlı)—————72.3%69.8%————————————————————————————————————————
MATEMATİK
AIME 2025———————99.8%——————————————98.1%95.0%———————————96.0%———————91.7%———
AIME 2026——————95.8%96.7%95.1%——————97.5%99.2%—————98.3%91.7%—95.8%—99.2%95.8%——95.3%——95.8%94.2%———————————
USAMO 2026———97.6%———66.2%————————95.2%—————74.4%————————————————————————
HMMT 2026 Feb———————96.2%————————97.7%—————94.7%————92.5%———87.8%——95.2%————————————
IMOAnswerBench———————75.3%————————91.4%—————81.0%————91.0%———83.8%——89.8%————————————
Apex———————34.5%————————54.1%—————60.9%18.4%——————————38.3%————————————
Apex Shortlist———————85.9%————————78.1%—————89.1%———————————90.2%————————————
ArxivMath——78.5%68.7%—71.8%—————————71.5%——————64.8%————————————————————————
RiemannBench——55.0%43.0%—34.0%—————————————————————————————————————————
UZUN BAĞLAM (1M Token)
MRCR 1M———————92.9%—————————————26.6%76.3%———————————83.5%————————————
CorpusQA 1M———————71.7%——————————————53.8%———————————62.0%————————————
GraphWalks BFS 256K——91.1%85.7%—85.9%76.9%38.7%———————73.7%21.4%——————————————————————————————
GraphWalks Parents 256K——99.96%99.9%—99.3%93.6%————————90.1%———————————————————————————————
EYLEMCİ / BİLGİSAYAR KULLANIMI
▸ Tarayıcı & Bilgisayar Kullanımı
BrowseComp——88.0%87.9%90.8%84.3%79.8%83.7%——84.7%74.7%———84.4%82.7%—————85.9%59.2%——————————83.4%——83.5%—————————
OSWorld-Verified——85.0%85.4%—83.4%82.8%72.7%66.3%—81.2%72.5%———78.7%75.0%————78.4%76.2%——————————————70.1%————————80.8%
ScreenSpot-Pro (araçsız)—————82.3%79.5%————————————————————————————————————————
ScreenSpot-Pro (araçlı)—————87.9%87.6%————————————————————————————————————————
Automation Bench——17.4%—26.0%15.5%9.9%———13.5%————12.9%—————14.5%9.6%————————————————————————
▸ Araç & Protokol Kullanımı
MCP-Atlas————85.8%82.2%79.1%75.8%62.3%——61.3%———75.3%68.1%————83.6%78.2%54.1%———76.8%———74.1%——73.6%——74.2%————————88.1%
Toolathlon———————47.2%———————55.6%54.6%————56.5%48.8%———————————51.8%————————————
τ2-bench (Retail)———————91.9%88.9%——91.7%——————————90.8%85.3%———————————————————————
τ2-bench (Telecom)———————99.3%98.2%——97.9%———98.0%98.9%—————99.3%98.0%———————————————————————
▸ Alan Eylemcileri (Finans/Hukuk/Ofis)
Finance Agent v1.1——————64.4%60.1%———————60.0%61.5%—————59.7%————————————————————————
Finance Agent v2—————53.9%51.5%————————51.8%—————57.9%43.0%———————————————————————57.2%
Legal Agent (Harvey seti)——13.3%——10.4%————5.8%————2.1%—————0.8%0.0%————————————————————————
Legal Agent (açık set)——16.9%13.4%—9.6%————8.9%————————————————————————————————————
OfficeQA Pro——57.9%——48.1%—————————52.6%——————18.1%————————————————————————
Vending-Bench 2 ($)——$5.680—$11.182$5.787$10.937$8.018$4.967—$6.378$7.204——$9.619$7.524$6.144————$5.396$911$5.478$5.165$6.205—$8.314$5.634——$5.115——$3.285$1.034—$2.158——$3.887—$4.663————
▸ Belge & İktisadi Görevler
HLE (araçlı)——64.5%64.7%64.7%57.9%54.7%53.3%——57.4%49.0%———52.2%58.7%—————51.4%————————50.6%——48.2%————————44.4%———
GDPval-AA (Elo)——1932——1890175316191450——1676———17691674————165613141184—1481—15241535—13001352——155411974091670———10981168990———
AA-Briefcase (Elo)—16661538—165113241260———13621071—156814831143——12081124959873466—1504820152412439691398918——1273930——1103—15541293764——15621354868
GDP.pdf (görsel)——29.8%——22.5%—————————24.9%——————16.7%————————————————————————
SİBER GÜVENLİK
CyberGym———83.1%——73.1%73.8%———————81.8%66.3%——————————————————————————————
ExploitBench——78.0%69.0%—40.0%—————————34.0%———————————————————————————————