News-Tracker

Modeller

AI & LLM · daglig briefing

Ændringer siden forrige snapshot

Ny i indekset

  • Claude Opus 4.8 (Adaptive Reasoning, Max Effort) er ny i Artificial Analysis' indeks (intelligence 42, Anthropic).
  • K2 Horizon MoVA 36B A4B er ny i Artificial Analysis' indeks (intelligence 25.7, MBZUAI Institute of Foundation Models).
  • K2 Horizon 7B er ny i Artificial Analysis' indeks (intelligence 21, MBZUAI Institute of Foundation Models).
  • K2 Horizon 3.7B er ny i Artificial Analysis' indeks (intelligence 16.2, MBZUAI Institute of Foundation Models).
  • K2 Horizon 0.9B er ny i Artificial Analysis' indeks (intelligence 3, MBZUAI Institute of Foundation Models).

Score ændret

  • GLM-5.2 (max): intelligence 38.6 → 34 (-4.6).
  • Ling 3.0 Flash: intelligence 24.9 → 20.6 (-4.3).
  • Kimi K2.6: intelligence 31.3 → 27.5 (-3.8).

Ny i HF-trending

  • internlm/Atria-Dawn-Preview er ny i Hugging Faces trending (score 61).
  • IFM/K2-Horizon-7B er ny i Hugging Faces trending (score 50).

Intelligence Index — Artificial Analysis

#ModelIntel.Agentic CreatorUdgivet
1 Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) 5 effort-varianter 53.4 58 Anthropic 2026-09-01
2 GPT-6 Astra (max) 5 effort-varianter 52.8 51.5 OpenAI 2026-09-03
3 Claude Opus 5 (Adaptive Reasoning, Max Effort) 5 effort-varianter 50.7 56.2 Anthropic 2026-07-24
4 Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) 49.7 51 Anthropic 2026-06-09
5 Muse Spark 1.3 (max) 2 effort-varianter 48.2 55.7 Meta 2026-09-02
6 GPT-5.6 Sol (max) 6 effort-varianter 47.1 50.5 OpenAI 2026-07-09
7 GLM-5.3 (max) ⬇ hf.co/zai-org/GLM-5.3 44.9 53.4 Z AI 2026-08-18
8 Grok 4.6 (high) 4 effort-varianter 44.4 53.4 SpaceXAI 2026-08-12
9 Kimi K3 (max) 2 effort-varianter ⬇ hf.co/moonshotai/Kimi-K3 43.8 50.6 Kimi 2026-07-16
10 GPT-5.6 Terra (max) 6 effort-varianter 42.3 43.7 OpenAI 2026-07-09
11 Claude Opus 4.8 (Adaptive Reasoning, Max Effort) 42 42.6 Anthropic 2026-05-28
12 GLM-5.3-Flash ⬇ hf.co/zai-org/GLM-5.3-Flash 41.9 51.2 Z AI 2026-08-26
13 Gemini 3.8 Flash (high) 3 effort-varianter 41.2 41.1 Google 2026-09-02
14 Claude Opus 4.7 (Adaptive Reasoning, Max Effort) 2 effort-varianter 40.7 39.5 Anthropic 2026-04-16
15 Qwen3.8 Max 40.3 49.6 Alibaba 2026-08-03
16 Qwen3.8 2.4T A95B ⬇ hf.co/Qwen/Qwen3.8-2.4T-A95B 40 50.4 Alibaba 2026-08-12
17 Qwen3.8-Flash-Next ⬇ hf.co/Qwen/Qwen3.8-Flash-Next 39.9 53.9 Alibaba 2026-08-26
18 Muse Spark 1.2 (xhigh) 39.8 44 Meta 2026-08-05
19 Gemini 3.7 Flash (medium) 3 effort-varianter 39.6 Google 2026-08-13
20 DeepSeek V4.1 Flash (Reasoning, Max Effort) ⬇ hf.co/deepseek-ai/DeepSeek-V4.1-Flash 39.5 DeepSeek 2026-09-10
21 Grok 4.5 (high) 39.1 42.1 SpaceXAI 2026-07-08
22 GPT-5.4 (xhigh) 3 effort-varianter 39 OpenAI 2026-03-05
23 GPT-5.5 (xhigh) 5 effort-varianter 38.6 37.3 OpenAI 2026-04-23
24 Claude Sonnet 5 (Adaptive Reasoning, Max Effort) 6 effort-varianter 38.4 44.3 Anthropic 2026-06-30
25 GPT-5.6 Luna (max) 6 effort-varianter 37.5 42.7 OpenAI 2026-07-09
26 DeepSeek V4 Pro 0813 (Reasoning, Max Effort) ⬇ hf.co/deepseek-ai/DeepSeek-V4-Pro 36.3 42.3 DeepSeek 2026-08-13
27 Agnes 3.0 Flash ⬇ hf.co/Agnes-AI/Agnes-3.0-Flash 35.5 Sapiens AI 2026-09-11
28 Agnes 2.5 Pro Beta 35.2 Sapiens AI 2026-08-26
29 DeepSeek V4 Flash Vision (Reasoning, Max Effort) 35 48.1 DeepSeek 2026-08-21
30 DeepSeek V4 Flash 0731 (Reasoning, Max Effort) ⬇ hf.co/deepseek-ai/DeepSeek-V4-Flash-0731 34.5 41.7 DeepSeek 2026-07-31
31 Gemini 3.6 Flash (high) 34.3 30.2 Google 2026-07-21
32 Muse Spark 1.1 (xhigh) 34.3 27.5 Meta 2026-07-09
33 GLM-5.2 (max) 2 effort-varianter ⬇ hf.co/zai-org/GLM-5.2 34 39.4 Z AI 2026-06-16
34 Qwen3.8 27B (xhigh) 4 effort-varianter ⬇ hf.co/Qwen/Qwen3.8-27B 33.9 46.5 Alibaba 2026-08-14
35 Gemini 3.5 Flash (medium) 3 effort-varianter 33.6 Google 2026-05-19
36 Motif 3 2 effort-varianter ⬇ hf.co/Motif-Technologies/Motif-3 33.6 Motif Technologies 2026-08-12
37 GPT-5.3 Codex (xhigh) 32.5 OpenAI 2026-02-05
38 Claude Opus 4.6 (Adaptive Reasoning, Max Effort) 2 effort-varianter 31.9 Anthropic 2026-02-05
39 Muse Spark 31.3 Meta 2026-04-08
40 DeepSeek V4 Pro (Reasoning, Max Effort) 3 effort-varianter ⬇ hf.co/deepseek-ai/DeepSeek-V4-Pro 30.9 27.7 DeepSeek 2026-04-24

Hugging Face — trending familier

#Rod-repoTrendingLikes DownloadsParamsAlder
1 deepseek-ai/DeepSeek-V4.1-Flash +2 derivater 2180 2471 288414 763.2B 5 d
2 Qwen/Qwen3.6-35B-A3B +1 derivater 1784 2093 8109 34.7B 6 d
3 openbmb/MiniCPM5-2B +1 derivater 898 1399 206774 2.5B 9 d
4 nex-agi/Nex-N2.5-mini 620 781 4543 35.1B 7 d
5 Qwen/Qwen3.8-27B +12 derivater 574 15153 7703400 27.8B 41 d
6 nex-agi/Nex-N2.5-Pro 485 634 30489 396.8B 7 d
7 m-a-p/YuE2-3B 454 478 5186 3.6B 5 d
8 XHToken/Spark-X2.5-4B-Base +2 derivater 377 1176 24084 4.1B 22 d
9 Lightricks/LTX-2.5 373 3877 1559653 54 d
10 nreimers/MiniLM-L6-H384-uncased +1 derivater 241 5970 252806720 22.7M 1657 d
11 tencent/AuK +1 derivater 216 221 1928 28 d
12 MiniMaxAI/MiniMax-H3 +6 derivater 211 5292 4827156 33.1B 49 d
13 openai-community/gpt2 205 4077 15182177 137.0M 1657 d
14 Qwen/Qwen3.8-Flash-Next +4 derivater 197 5230 645881 180.0B 22 d
15 google-bert/bert-base-uncased 190 3337 46435111 110.1M 1657 d

Params er safetensors.total, altså lagrede vægte inklusive kvantiserede tensorer — ikke et mål for hvad modellen kræver af hardware.