AI Model Daily - 2026-05-31 | QName AI

30-Second Summary

No new general flagship LLM release was visible from OpenAI, Anthropic, Google DeepMind, Meta, Mistral, xAI, Qwen, or other major accounts in the last 24 hours; today's focus shifted to video model rankings, API distribution, and product availability signals.
Arena's official account confirmed inside the window that wan2.7-t2v reached #3 on the Text-to-Video Arena; the public leaderboard still shows Bytedance dreamina-seedance-2.0-720p at #1, happyhorse-1.0 at #2, and wan2.7-t2v at #3.
@xiaohu reported Koji, an AI tutor positioned as a "graphical private tutor" / screen-guided learning product. For now this is a media/community lead rather than a foundation model release, and no verifiable model card was found.
Gemini Omni, Runway API's Seedance 2.0, and Cohere Command A+ translation capability were all visible in this X pass, but their original post times landed a few hours before the strict 24-hour window; they remain high-priority carryover signals and are not counted as today's first releases.
Leaderboards were broadly stable: Artificial Analysis still has Claude Opus 4.8 leading LLM Intelligence and GDPval-AA; LMArena Text remains dated 2026-05-28 and still does not show Opus 4.8; Artificial Analysis image/video leaderboards did not show a new #1 replacement.

Must Watch

1. Arena: Wan2.7-t2v Reached #3 on Text-to-Video Arena

Event type: Arena appearance + X first signal.
First/official time: @arena posted at 2026-05-30 09:25 HKT; the public Text-to-Video Arena page is dated 2026-05-29.
Capability change: Arena said Wan2.7-t2v-2026-04-25 is #3 on the Text-to-Video Arena. The public leaderboard in labs view shows Bytedance dreamina-seedance-2.0-720p at 1463+/-9 in #1, Alibaba-ATH happyhorse-1.0 at 1440+/-8 in #2, Alibaba wan2.7-t2v at 1385+/-21 in #3, Google veo-3.1-audio-1080p at 1370+/-11 in #4, OpenAI sora-2-pro at 1367+/-8 in #5, and xAI grok-imagine-video-720p at 1359+/-7 in #6.
Availability: Arena leaderboard / Battle Mode comparisons are available; official API, model card, pricing, and exact version identity still need confirmation from Wan/Alibaba official sources.
Impact: today's most useful signal is a video-model ranking move, not a text-model release. Wan2.7 entering the top three of Arena's video leaderboard suggests Alibaba's video line is close to the Seedance / HappyHorse top tier in open arena evaluation, but the score interval is wide and the model version comes from Arena naming, so official model-card and availability checks remain necessary.
Sources: Arena Text-to-Video Leaderboard, Arena X

2. Runway: Seedance 2.0 API Distribution Matters for Video Workflows

Event type: official product update + X-visible distribution.
First/official time: Runway API changelog shows 2026-05-28; @runwayml had a visible promotion post at 2026-05-30 04:10 HKT, slightly before this strict 24-hour window.
Capability change: Runway API changelog confirms Seedance 2.0 is available through the Runway API, supporting text-to-video, image-to-video, and video-to-video, with keyframe control, reference images, reference videos, generated audio, and generation lengths of 4-15s.
Availability: Runway API, with seedance2 used as the model parameter in examples; docs show text-to-video, image-to-video, video-to-video, and audio-reference examples.
Impact: Seedance 2.0 was not first released today, but its Runway API distribution has high product-engineering value. It turns a high-ranking video model into a developer-accessible capability. Next checks should cover latency, failure rate, audio-reference consistency, generation cost, and commercial-use constraints.
Sources: Runway API changelog, Runway Seedance guide, Runway X

3. Google Gemini Omni: User Availability Keeps Expanding, API Still Unconfirmed

Event type: official availability signal + X first-signal tracking.
First/official time: Google's I/O subscription update article was published on 2026-05-19; @GeminiApp had multiple visible Gemini Omni demo posts at 2026-05-30 05:30 HKT, slightly before this strict 24-hour window.
Capability change: Google says Gemini Omni is available globally to AI Plus, Pro, and Ultra users, can create and edit videos from text, image, and video inputs, and supports conversational iteration in the Gemini app / Google Flow. Google also emphasizes character consistency, identity, and voice consistency across scenes.
Availability: Gemini app and Google Flow; visible @GeminiApp posts said AI Plus, Pro, and Ultra subscribers can use it at gemini.google and in the app. No clear opening was found for Gemini API, AI Studio, Vertex, or Gemini Enterprise.
Impact: Gemini Omni remains a must-track item this week, but today was not an API launch. It should be classified as a high-value user-side video model rather than a developer-integrable model. Key checks: quota consumption, whether failed requests are charged, video input length, character/voice consistency, and API roadmap.
Sources: Google AI subscriptions, Gemini Drops, Gemini X

4. XiaoHu: Koji AI Tutor Is a Product Signal, Not a Model Launch

Event type: media report + product lead requiring verification.
X time: @xiaohu posted the Koji introduction at 2026-05-30 10:14 HKT.
Capability change: XiaoHu described Koji as an AI tutor that "refuses to give the answer" and as a "graphical private tutor"; the core idea is seeing the screen, highlighting/annotating content, and guiding students in real time instead of directly giving answers.
Availability: the XiaoHu post linked to xiaohu.ai/c/xiaohu-ai/koji-ai; public retrieval currently showed the article page as "post was not found", so this report records only the visible X content.
Impact: this kind of "screen understanding + pedagogy + real-time guidance" product is worth tracking because it points toward the same visual tutor surface as Gemini Live, ChatGPT learning features, and Claude agent workflows. But no official model name, model card, API, or independent availability entry was found, so it should not be listed as a foundation-model launch.
Sources: XiaoHu X, XiaoHu.AI search result

Arena Watch

LMArena Text Overall: the page remains dated 2026-05-28, with 6,532,755 votes and 360 models. The top entries are still claude-opus-4-6-thinking 1502+/-4, claude-opus-4-7-thinking 1500+/-5, claude-opus-4-6 1498+/-4, claude-opus-4-7 1494+/-5, muse-spark 1489+/-6, gemini-3.1-pro-preview 1487+/-4, gemini-3-pro 1486+/-4, and gpt-5.5-high 1482+/-6. Opus 4.8 still does not have an official visible score there.
Arena Code / WebDev: the page is dated 2026-05-24, with 328,594 votes and 81 models. The top entries are claude-opus-4-7-thinking 1567+10/-10, claude-opus-4-7 1562+10/-10, claude-opus-4-6-thinking 1542+8/-8, qwen3.7-max-20260517 1541+16/-16, and claude-opus-4-6 1538+7/-7. Opus 4.8 has appeared in Arena X Battle Mode / test videos, but the Code leaderboard still does not show Opus 4.8.
Arena Text-to-Video: the page is dated 2026-05-29, with 452,799 votes and 17 labs. Labs view shows Bytedance dreamina-seedance-2.0-720p 1463+/-9 in #1, Alibaba-ATH happyhorse-1.0 1440+/-8 in #2, Alibaba wan2.7-t2v 1385+/-21 in #3, Google veo-3.1-audio-1080p 1370+/-11 in #4, OpenAI sora-2-pro 1367+/-8 in #5, and xAI grok-imagine-video-720p 1359+/-7 in #6.
Artificial Analysis LLM: Claude Opus 4.8 (max) has Index 61, cost $4.10/M, speed 59 tokens/s; GPT-5.5 (xhigh) has 60, $4.35/M, 52 tokens/s; GPT-5.5 (high) has 59; Claude Opus 4.7 (max) has 57; Gemini 3.1 Pro Preview has 57, $1.74/M, 114 tokens/s; Qwen3.7 Max has 57, $1.43/M, 189 tokens/s; Gemini 3.5 Flash has 55, $1.31/M, 183 tokens/s.
GDPval-AA: Claude Opus 4.8 (Adaptive Reasoning, Max Effort) remains first with Elo 1890 and CI -34/+35; GPT-5.5 (xhigh) is 1769, GPT-5.5 (high) 1755, Claude Opus 4.7 (max) 1753, Gemini 3.5 Flash (medium) 1658, Gemini 3.5 Flash (high) 1656, and Qwen3.7 Max 1548.
ITBench-AA: Opus 4.8 still does not appear; the page summary shows Claude Opus 4.7 (max) first at 46.7%, GPT-5.5 (xhigh) second at 45.8%, and Qwen3.7 Max third at 42.5%. Real SRE / enterprise IT agent tasks still have not crossed 50%.
Artificial Analysis Text to Image: GPT Image 2 (high) leads with Elo 1338, followed by GPT Image 1.5 (high) 1268, Nano Banana 2 (Gemini 3.1 Flash Image Preview) 1260, Nano Banana Pro (Gemini 3 Pro Image) 1219, and Recraft V4.1 Utility Pro 1205.
Artificial Analysis Image Editing: GPT Image 1.5 (high) leads at 1260, followed by GPT Image 2 (high) 1252, Nano Banana Pro 1241, Nano Banana 2 Preview 1240, and grok-imagine-image-quality 1230.
Artificial Analysis Text to Video: with audio, Dreamina Seedance 2.0 720p remains first with Elo 1215, HappyHorse-1.0 second at 1212, Kling 3.0 1080p (Pro) 1103, Kling 3.0 Omni 1080p (Pro) 1099, and Veo 3.1 1094; without audio, HappyHorse-1.0 remains first with Elo 1358, followed by Dreamina Seedance 2.0 720p 1272, Kling 3.0 1080p (Pro) 1250, Kling 3.0 Omni 1080p (Pro) 1235, and grok-imagine-video 1234.
Sources: LMArena Text, Arena Code, Arena Text-to-Video, Artificial Analysis LLM, GDPval-AA, ITBench-AA, Text to Image, Image Editing, Text to Video

X Watch

@arena: inside the window, posted Wan2.7-t2v-2026-04-25 is #3 on the Text-to-Video Arena; this is today's clearest new Arena signal. The post quotes @Alibaba_Wan's Wan2.7-Video introduction, but the model's first release itself was not today.
@xiaohu: inside the window, posted the Koji AI tutor introduction, emphasizing screen understanding, highlighting/annotation, and real-time guidance; this is a media/community report and AI product lead, not a model release. XiaoHu's AI daily / academy pages remain auxiliary sources, but the Koji article page is not publicly readable right now.
@GeminiApp: visible Gemini Omni demo posts said it can understand video input, apply physics effects, and generate new motion; another post said Gemini Omni is available to Google AI Plus, Pro, and Ultra subscribers. These X posts were slightly before the strict window and are treated as high-priority carryover.
@runwayml: visible Runway API promo post said the API aggregates models such as Seedance 2.0 and GPT Image 2; the timing was slightly before the strict window. Official changelog confirms Seedance 2.0 API distribution.
@cohere: visible Command A+ machine-translation capability post said it leads Mistral Medium 3.5, DeepSeek, OpenAI gpt-oss, Claude Opus 4.6, and others on WMT24++ / xCOMET-XL; the timing was slightly before the strict window, so it is logged as model-capability propagation rather than a new release.
@OpenAI: no new general model release was visible inside this window; the latest visible signals were still the Terence Tao AI math event, Codex Windows Computer Use, and Rosalind Biodefense from the prior cycle.
@AnthropicAI / @ClaudeAI: no new model release was visible inside this window; latest visible signals still centered on Claude Opus 4.8, Dynamic Workflows, and financing/revenue diffusion.
@ArtificialAnlys: no new leaderboard post was visible inside this window; latest visible signals still included Opus 4.8 Intelligence Index 61.4, GDPval-AA leadership, and grok-imagine-image-quality entering the top tier of image leaderboards.
@xai: no new model release was visible inside this window; latest visible signals were still grok-build-0.1 API public beta, OpenRouter/Vercel AI Gateway/agent IDE entry points, and Grok Build 0.2.7.
@GoogleDeepMind: no new model release was visible inside this window; latest visible signal was still reposting Nano Banana 2 / Nano Banana Pro GA.
@Alibaba_Qwen: no new model release was visible inside this window; latest visible signals still centered on Qwen3.7-Max ITBench-AA and Qwen3.5 TokenSpeed.
@MistralAI: no new model release was visible inside this window; latest visible signals were still AI Now Summit, Mistral Vibe, and industrial/physical AI solutions.
@AIatMeta: no new model release was visible inside this window; latest model-related content remained the Muse Spark pinned / older posts.
@LumaLabsAI: visible Luma Agents product post about turning assets into promos was slightly before the window; no new model first release.
@Kling_ai: no new model release was visible inside this window; latest visible signals were still Cannes / AI film workflow cases.

To Confirm

wan2.7-t2v: Arena confirmed it reached #3 on the Text-to-Video Arena, but official model card, API name, pricing, available regions, and whether it exactly maps to Wan2.7-t2v-2026-04-25 still need confirmation.
happyhorse-1.0: Arena and Artificial Analysis video leaderboards remain strong, but no public official model card or API was found; Artificial Analysis still shows API Coming soon.
Gemini Omni: official sources confirm availability through Gemini app / Google Flow and AI Plus/Pro/Ultra users, but no clear opening was found for Gemini API / AI Studio / Vertex / Gemini Enterprise; developer entry points and quota consumption need continued tracking.
Runway Seedance 2.0 API: official docs confirm seedance2, but text/image/video/audio-reference stability, generation-length limits, failure rate, pricing, content moderation, and commercial terms still need hands-on testing.
Koji AI tutor: currently only the XiaoHu X report and a publicly unreadable article page were found; official site, product entry point, model architecture, privacy boundary, and real availability need second-source verification.
Claude Opus 4.8: Artificial Analysis LLM/GDPval already show it leading, but LMArena Text, Arena Code, and ITBench-AA still do not have complete official scores; continue watching Battle Mode conversion into leaderboard entries.
Nano Banana 2 / Pro: Google Cloud GA is confirmed, but Artificial Analysis still shows Nano Banana 2 Preview naming; continue watching whether it is replaced by GA entries and whether scores update.
Cohere Command A+: X-side translation comparison is visible, but official technical report / evaluation setup needs confirmation, especially WMT24++ / xCOMET-XL languages, baselines, and whether comparisons used the same prompt and cost conditions.
AlibabaCloud X: @AlibabaCloud resolved to an anomalous title in this pass and was not used as a valid signal; future checks should prioritize Qwen / Model Studio / Wan official accounts and official docs.