QName AI logoQName AI
All AI model reports
Collected in Hong Kong time

AI Model Daily - 2026-06-29

2026-06-29 AI model news: Grok 4.5 surfaced through Elon Musk's X post as a SpaceX and Tesla private beta; Sakana Fugu Ultra remains available through Vercel AI Gateway; HappyHorse 1.1 stays high on video arenas; no fresh official releases from OpenAI, Anthropic, Google, Qwen, Mistral, Meta, Cohere, Runway, Luma, Kling, Midjourney, or Microsoft AI in this window.

30-Second Summary

  • The only clear new model-level signal today came from Elon Musk's X post: Grok 4.5 is described as based on a 1.5T V9 foundation model, with Cursor data added during supplemental training, and is now in private beta at SpaceX and Tesla. Early evaluations were described as close to, and possibly above, Opus. There is still no xAI official blog post, model card, or API ID, so classify this as "X-first + pending verification."
  • @xiaohu amplified the same Grok 4.5 signal in Chinese within the window, making it the most important Chinese-side X lead today. The public Xiaohu AI Daily page is still on its 2026-06-27 issue, so it is useful as background but should not replace official sources.
  • Sakana Fugu Ultra was promoted again by @SakanaAILabs as available on Vercel AI Gateway. This is a continuing availability confirmation, not a new launch today; the model ID is sakana/fugu-ultra, and the original Vercel changelog date remains 2026-06-22.
  • Arena / Artificial Analysis did not show a new model-entry post or leaderboard-change announcement in this window. HappyHorse 1.1 remains the key video-board item: Artificial Analysis currently has it at #2 for T2V with audio and #2 for T2V without audio.
  • OpenAI, Anthropic, Google DeepMind/Gemini, Qwen, Mistral, Meta, Cohere, Runway, Luma, Kling, Midjourney, and Microsoft AI had no official signal in this window that should be upgraded to a new model release. GPT-5.6, HappyHorse 1.1, and Runway seedance2_mini remain carry-over watch items.

Must Watch

1. Grok 4.5: Elon Musk's private-beta signal, with no public API yet

  • Event type: X-first signal + pending verification.
  • First/official/availability time: Elon Musk posted at 2026-06-28T10:50Z; @xiaohu posted a Chinese relay at 2026-06-28T13:12Z. No xAI official blog, docs, or API model page was found in this run.
  • Model/function names: Grok 4.5, 1.5T V9 foundation model, Cursor supplemental training, Grok Build harness.
  • Capability change: Elon said Grok 4.5 is based on the 1.5T V9 foundation model and had Cursor data added during supplemental training. Early evaluations were described as close to, and possibly exceeding, Opus. He also said RL is still significantly improving the model and that the Grok Build harness continues to improve.
  • Availability: SpaceX and Tesla private beta only; no public xAI API ID, Grok product entry, model card, or reproducible third-party test yet.
  • Impact: this is the highest-priority signal today, but it must be separated by evidence quality. The source is the founder's X account, stronger than a generic rumor; however, the Opus comparison, Cursor-data details, private-beta scope, rollout plan, and evaluation methodology all lack official documentation and third-party validation. Keep watching xAI docs, Grok product pages, API model lists, and whether grok-4.5 appears on LMArena or Artificial Analysis.
  • Sources: @elonmusk Grok 4.5, @xiaohu relay, xAI docs

2. Fugu Ultra: Sakana keeps promoting Vercel AI Gateway availability

  • Event type: official/platform availability change.
  • First/official/availability time: the Vercel changelog is dated 2026-06-22; @SakanaAILabs promoted Vercel AI Gateway availability again at 2026-06-28T18:30Z.
  • Model/function names: Sakana Fugu Ultra, sakana/fugu-ultra.
  • Capability change: Vercel describes Fugu Ultra as a system built on multiple publicly accessible frontier models. It dynamically routes a request to 1-3 agents and aggregates their outputs, positioning the endpoint for reasoning, scientific benchmarks, coding, and other complex tasks.
  • Availability: Vercel AI Gateway; in AI SDK, use model: 'sakana/fugu-ultra'.
  • Impact: this is not a new model launch today, but it is an official availability reminder within the window. Fugu Ultra deserves hands-on testing because it packages model routing / multi-agent aggregation behind one commercial model entry. The main risks to test are underlying model pool, latency, cost, data boundaries, and reproducibility.
  • Sources: @SakanaAILabs, Vercel Changelog: Sakana Fugu Ultra

3. HappyHorse 1.1: still the key video-arena watch item, but not a new launch today

  • Event type: arena appearance + continuing leaderboard watch.
  • First/official/availability time: @arena posted HappyHorse 1.1's arena entry at 2026-06-26T21:56Z, so it is not a new first appearance in today's window. Artificial Analysis still lists HappyHorse-1.1 among video models added in the last month.
  • Model/function names: HappyHorse 1.1, HappyHorse 1.0, Video Arena, Artificial Analysis Video Arena.
  • Capability change: Arena covers Text-to-Video, Image-to-Video, and Video Edit. On Artificial Analysis, T2V with audio currently has Dreamina Seedance 2.0 720p at Elo 1219 in #1, HappyHorse 1.1 at Elo 1152 in #2, and HappyHorse 1.0 at Elo 1123 in #3. T2V without audio currently has HappyHorse 1.0 at Elo 1290 in #1, HappyHorse 1.1 at Elo 1285 in #2, and Dreamina Seedance 2.0 720p at Elo 1274 in #3.
  • Availability: Arena / Artificial Analysis voting and comparison are available. Runway API changelog already lists happyhorse_1_0, but this run still did not find a happyhorse_1_1 API ID.
  • Impact: HappyHorse 1.1 is in the observable-leaderboard stage, not the stable-API stage. Continue tracking Video Edit rank, audio-dimension performance, and whether an official model card or API ID appears.
  • Sources: @arena HappyHorse 1.1, Artificial Analysis Text to Video, Arena Text-to-Video, Runway API changelog

4. GPT-5.6: still the highest-priority recent official release, with no new official move today

  • Event type: official release + X-first signal + continuing watch.
  • First/official/availability time: the first @OpenAI post was at 2026-06-26T17:10Z; OpenAI's official release page is dated 2026-06-26. No new OpenAI GPT-5.6 model-release post appeared in this window.
  • Model/function names: GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, max reasoning effort, ultra mode.
  • Capability change: OpenAI positions Sol as the GPT-5.6 flagship, emphasizing coding, biology, cybersecurity, and agentic workflows; Terra targets balanced everyday cost/performance; Luna targets high-speed low-cost usage.
  • Availability: during limited preview, access starts through API and Codex for trusted partners and organizations. Broader ChatGPT, Codex, and API rollout timing still needs verification.
  • Impact: there is no new update today, but GPT-5.6 remains the main model-capability and governance thread for the week. Next tests should prioritize Codex/API quotas, whether max/ultra improves long-horizon agent workflows, and whether Cyber/Bio safety enforcement affects legitimate defensive work.
  • Sources: OpenAI: Previewing GPT-5.6 Sol, GPT-5.6 Preview System Card, @OpenAI GPT-5.6

5. OpenAI Devs: Codex scheduled-work article amplified by the official developer account

  • Event type: X-first/official product signal; not a model release.
  • First/official/availability time: @OpenAIDevs reposted Jason Liu's article about Codex scheduled work at 2026-06-28T07:39Z.
  • Model/function names: Codex scheduled work, background tasks, recurring work.
  • Capability change: this is not a new model, but it reinforces OpenAI's direction of moving Codex from an interactive coding agent toward scheduled and automatically executed workflows.
  • Availability: the X article reposted by OpenAI Devs; concrete Codex availability should be checked in Codex product UI and official documentation.
  • Impact: for automation workflows such as this daily report, Codex scheduled work is worth tracking. It should not be classified as a model-capability update today.
  • Sources: @OpenAIDevs repost, Jason Liu original post

Arena Watch

  • LMArena / Arena Video: no new @arena model-entry post or leaderboard-change announcement was seen in this window. HappyHorse 1.1 remains the main observation target after its 2026-06-26 entry, and should stay classified as "arena appearance" until scores stabilize.
  • Artificial Analysis Text-to-Video with audio: Dreamina Seedance 2.0 720p Elo 1219 #1, HappyHorse 1.1 Elo 1152 #2, HappyHorse 1.0 Elo 1123 #3, Kling 3.0 1080p Pro #4, SkyReels V4 #5.
  • Artificial Analysis Text-to-Video without audio: HappyHorse 1.0 Elo 1290 #1, HappyHorse 1.1 Elo 1285 #2, Dreamina Seedance 2.0 720p Elo 1274 #3, Kling 3.0 1080p Pro #4, Kling 3.0 Omni 1080p Pro #5.
  • gemini-omni-flash: still a high-priority arena/anonymous or not-fully-officialized signal. This run did not find a Google / Gemini official model card, API ID, or release page confirming it.
  • Grok 4.5: no LMArena or Artificial Analysis reproducible leaderboard entry was seen in this window. For now, it should be written as a founder X signal and private-beta claim.
  • Sources: Arena Text-to-Video, Arena Video Editing, Artificial Analysis Text to Video, @arena HappyHorse 1.1

X Watch

  • @elonmusk: at 2026-06-28T10:50Z, posted the Grok 4.5 private-beta signal, describing the model as based on a 1.5T V9 foundation model, with Cursor data added during supplemental training, and now in private beta at SpaceX / Tesla. Classification: X-first + pending verification.
  • @xiaohu: at 2026-06-28T13:12Z, relayed the Grok 4.5 signal. This is the most important Chinese-side X lead in the window. The public Xiaohu AI Daily page is still on its 2026-06-27 issue, so use it as background only. Classification: media/personal source + pending verification.
  • @SakanaAILabs: at 2026-06-28T18:30Z, again promoted Fugu Ultra's availability on Vercel AI Gateway. Classification: official/platform availability change.
  • @OpenAIDevs: at 2026-06-28T07:39Z, reposted an article about Codex scheduled work. Classification: official product/workflow signal; not a model release.
  • @Kling_ai: at 2026-06-28T15:00Z, posted character-card / engagement content, with no new Kling model ID seen. Classification: product/activity signal; not a model launch.
  • @OpenAI: no new model post in this window; GPT-5.6 remains the 2026-06-26 official release. Classification: no new model launch.
  • @AnthropicAI / @claudeai: no new Claude model release in this window; Mythos 5 limited restoration remains a 2026-06-27 signal. Classification: no new model launch.
  • @arena / @ArtificialAnlys: no new model entry or leaderboard announcement in this window; HappyHorse 1.1 and Model Sets are carry-over items. Classification: continuing leaderboard watch.
  • @GoogleDeepMind / @GoogleAI / @GeminiApp: no official Gemini Omni, Nano Banana, or new Gemini model release was seen in this window. Classification: no new model launch.
  • @Alibaba_Qwen / Wan-related accounts: no official Qwen or Wan model card/API release was seen in this window; Wan Streamer remains primarily a previous-window Xiaohu signal. Classification: pending verification.
  • @MistralAI, @xai, @grok, @AIatMeta, @cohere, @runwayml, @LumaLabsAI, @midjourney, @MicrosoftAI: no official X signal in this window was strong enough to classify as a new model release. Classification: no new model launch.
  • Sources: @elonmusk Grok 4.5, @xiaohu, @SakanaAILabs, @OpenAIDevs, @Kling_ai

Pending Confirmation

  • Grok 4.5 officialization: wait for xAI blog, docs, API model list, Grok product entry, model card, system card, or reproducible third-party testing. The "possibly exceeding Opus" claim should not be written as verified.
  • Grok 4.5 data and compliance: how Cursor data was added during supplemental training, data authorization/privacy boundaries, and whether this affects coding capability or enterprise adoption need official explanation.
  • Grok 4.5 rollout path: outside the SpaceX / Tesla private beta, keep watching for Grok App, API, SuperGrok, Grok Build, or LMArena availability.
  • SpaceX monthly model plan: Elon mentioned new from-scratch models from SpaceX every month this year, but there is no formal roadmap, naming scheme, use-case description, or release mechanism yet.
  • Fugu Ultra reproducibility: Vercel makes it callable, but the underlying model pool, routing strategy, privacy/data retention, latency, cost, fallback behavior, and output stability still need testing.
  • HappyHorse 1.1 API: Arena / AA make it observable, but there is still no official happyhorse_1_1 API ID, pricing, model card, or commercial terms.
  • gemini-omni-flash: still needs a Google / Gemini official model card, API ID, pricing, or product entry.
  • Wan Streamer: needs a Wan / Alibaba official blog post, model card, paper, demo URL, or API documentation confirming real-time video-call capability and access scope.
  • GPT-5.6 rollout: OpenAI's coming-weeks ChatGPT, Codex, and API batches, regions, rate limits, real latency, and cost still need hands-on verification.
  • Runway / Kling / Luma video models: no new model launch appeared in this window, but continue watching whether video editing, character consistency, audio synchronization, and real-time interaction capabilities get formal APIs.