EN
AI Model Tracker · Nov 14, 2026 edition

AI Model Tracker Nov 14, 2026: What changed for mid-sized companies

Every 14 days we check which AI models lead public leaderboards. Since the Oct 14 issue, 9 of 36 use cases have changed, mainly in text, code, agents, image and video.

What changed

Text

Claude Fable 5.2 Max (released Nov 9) now leads the Artificial Analysis Intelligence Index with 65 points. Gemini 4 Carbon High follows with 60, and Claude Opus 5.5 Max with 59. In the Arena Text list, Gemini 4 Carbon High (released Nov 10) leads with 1526 points ahead of Gemini 4 Argon High with 1522; both values are provisional. Fable 5.2 is not listed there yet. The best open model on the index is Kimi K3.1 Pro at 46 points (weights public since Nov 4), ahead of DeepSeek V4.1 Flash 1017 at 33. Kimi uses its own custom license, which is not open source in the strict sense.

Coding

In the Arena WebDev list, Claude Fable 5.2 xHigh leads with 1838 points, ahead of Opus 5.5 Max with 1815, Sonnet 5.5 xHigh with 1772, and Gemini 4 Carbon High with 1768 (provisional). The best open model is Kimi K3.1 Pro Max at 1687, ahead of DeepSeek V4.1 Flash 1017 at 1664.

Agents and tool calling

Claude Fable 5.2 Max has taken first place in the Arena Agent list with a 30.48 percent net improvement, ahead of Opus 5.5 Max at 19.51 percent and Gemini 4 Carbon High at 16.90 percent (provisional). According to the Arena, the median cost per task is 8.02 US dollars for Fable 5.2 Max and 1.42 US dollars for Gemini 4 Carbon High. The best open model, Kimi K3.1 Pro Max, reaches 4.83 percent.

Image

In image generation, GPT Image 2.5 Sunburst remains in front with 1429 points. GPT Image 2.6 Sunburst Medium is now in 2nd place with 1406 (provisional). In image editing (single edit), GPT Image 2.5 Sunburst remains number 1 with 1519, and GPT Image 2.6 Sunburst Medium is newly in 3rd place (1476, provisional), ahead of MAI-Image-2.7 Preview with 1471.

Image understanding

A correction to the last issue: In the Vision list, Claude Opus 5.5 High is not in 2nd place but in 6th place with 1295 points. Claude Fable 5 remains number 1 there with 1309.

Video

Gemini Omni 1.2 Flash has been generally available since Nov 5. It generates clips up to 15 seconds and costs 0.07 US dollars per second in 720p. It now leads text to video with 1540 points and image to video with 1503; both values are provisional (lists dated Nov 8). In video to video, Wan 3.0 continues to lead with 1424 ahead of Seedance 2.5 with 1419; Omni 1.2 Flash is newly in 3rd place with 1396 (provisional, list dated Nov 10).

What this means for your business

  • For text and research, the top models are close to each other. A switch pays off mainly when price, data privacy or your tools support it.
  • For agents with tool access, Fable 5.2 is clearly in front, but its median cost per task is much higher than Gemini 4 Carbon. Plan for human approval when using agents.
  • If you do not want to hand data to US providers, you can run open models like Kimi K3.1 Pro yourself or with an EU provider. On agent tasks they are clearly behind the top.
  • We do not yet measure the cost per completed task in euros ourselves. Provisional rankings can still shift.

Unchanged: In the remaining 27 use cases, there has been no new number 1 since the Oct 14 issue. This also applies to music, voice, 3D, avatars and agent frameworks.

All 36 use cases

Top model and best open model per use case. Machine readable as JSON (CC BY 4.0).

Use caseTop modelsOpen models
TextClaude Fable 5.2 Max / Gemini 4 Carbon High / Gemini 4 Argon High / Claude Opus 5.5 MaxKimi K3.1 Pro / DeepSeek V4.1 Flash 1017 / GLM-5.3 Max / MiMo-V2.6-Pro
CodingClaude Fable 5.2 xHigh / Claude Opus 5.5 Max / Claude Sonnet 5.5 xHigh / Gemini 4 Carbon HighKimi K3.1 Pro Max / DeepSeek V4.1 Flash 1017 / Kimi K3 Max / MiMo-V2.6-Flash
Tool CallingClaude Fable 5.2 Max / Claude Opus 5.5 Max / Gemini 4 Carbon High / Claude Sonnet 5.5 MaxKimi K3.1 Pro Max / Inkling 2 Small / MiMo-V2.6-Pro / DeepSeek V4.1 Flash 1017
Image LabelingClaude Fable 5 / Qwen3.8 Max / Claude Opus 4.7 / Claude Opus 4.6GLM-5.3-Flash / Kimi K2.6 / Gemma 4 31B / Qwen3.5-397B
MusicAce Step 1.5 XL / Suno / Minimax Music 3HeartMuLa / Minimax Music 3
Avatar agentsSimli-
VoiceGemini 3.1 Flash TTS / ElevenLabs v3 / Grok Voice / Index TTS 2.5Qwen3TTS / VibeVoice
Voice to voiceElevenlabs / Heygen / Qwen3TTS / VibeVoice / WavTTS / Index TTS 2.5Scenema Audio / DramaBox / Lava SR / LongCat Audio / OmniVoice
Voice cloningElevenlabs / Tada / WavTTS / Index TTS 2.5Scenema Audio / DramaBox / IndexTTS2 / LongCat Audio / OmniVoice
Voice to textQwen3 ASR-
Realtime VoiceGrok Voice-
Text to ImageGPT Image 2.5 Sunburst / GPT Image 2.6 Sunburst Medium / GPT Image 2.5 Flare / GPT Image 2.6 Flare Medium-
Image editingGPT Image 2.5 Sunburst / GPT Image 2.5 Flare / GPT Image 2.6 Sunburst Medium / MAI-Image-2.7 Preview-
Text to VideoGemini Omni 1.2 Flash / Gemini Omni 1.1 Flash / Gemini Omni Flash / Flux 3 VideoMiniMax H3 (community) / LTX-2 / Wan 2.2 (Apache)
Image to VideoGemini Omni 1.2 Flash / Seedance 2.5 / MiniMax H3 / Gemini Omni 1.1 FlashMiniMax H3 (community) / Wan 2.2 (Apache) / LTX-2
Video to VideoWan 3.0 / Seedance 2.5 / Gemini Omni 1.2 Flash / Gemini Omni FlashWan-Animate-2 (Apache 2.0) / Wan-Animate-2-Lite / LTX-2
Video AvatarsLPM 1.0 / InfiniteTalk / Interactavatar / Flowact R1 / EditYourself / Live Portrait / omni show-
Video SegmentingLocateAnything / SAM3-
Lipsync videosOmniHuman Dreamina / EditYourselfJustDubIt / OmniSync / LatentSync / MultiTalk
Lipsync imagesOmniHuman Dreamina / Hedra / InfiniteTalk-
Consistent Characters / GNM GoogleGemini FlashFlux Kontext / Lynx
ReferencesRunway / Google Flow / Kling 3.0WAN Animate
AI UpscalingTopaz Labs / Sega (Image)PiD
UpframingTopaz Labs-
Video SoundFXPrismAudio / Elevenlabs / AudioX / Ace StudioPrismAudio
Agent frameworkCodex / Claude Cowork / Google Spark / PowerAutomate (Copilot Recorded)OpenClaw / Hermes Agent / n8n (selfhosted)
Scientific Agentic FrameworkRobin / Co-Scientist / Autoscientists-
Browser UseBrowser use / Auto GLM-
3D ModellingPixal3D / Hunyuan 3D 3 / MoCA / Trellis 2 / UltrashapeMoonlake
3D WorldMarble / Genie 3 / Mirage 2 / Trisplat-
3D AssetsArticraft / PhysX-Omni / CubePart-
ConciergeLorikeet-
SalesLusha / Clay / Apollo / Fiber-
Ad AgenciesHiggsfield / Superscale AI-
SAAS working with AIPosthog / Airtable-
Video EditorsPalmier / FTX Studio-

Method and sources

We evaluate public leaderboards and vendor announcements: arena.ai (standings up to Nov 11, 2026, video lists dated Nov 8 and Nov 10, 2026), the Artificial Analysis Intelligence Index, and publications from Anthropic, Google and Moonshot. Provisional values are marked. Next edition: Nov 28, 2026.

All issues