Top Chinese AI company on October 31, 2026? — Moonshot

- [2026-10-16] I could not open the resolution source, arena.ai/leaderboard/text. It returned a bot-check page. All the figures below come from third-party mirrors of Arena data, and they use Arena's default view (style control on). The contract uses the "Remove Style Control" view, so none of these numbers show that view directly. https://arena.ai/leaderboard/text - [2026-10 data] The llmrun.dev mirror says Kimi K3 is the top open model on the Arena text board at 1488.1, and #16 overall. The leader is Gemini 4 Argon High at 1525.4. Among open Chinese models, MiMo V2.6 Pro RL (Xiaomi) is next at 1480.4. After it come GLM 5.3 at 1478.4, GLM 5.2 at 1475.2 and DeepSeek V4.1 Flash at 1474.5. Both Xiaomi and Z.ai are mainland-China labs. The page is stamped "Data through Oct 2026". https://llmrun.dev/benchmark/lmarena-text - [2026-10-04] A dated snapshot from Sept 2 has Kimi K3 as the top Chinese entry, 12th at 1489 (±5). GLM-5.3 is 20th at 1482 and DeepSeek-V4-Pro-0813 is 52nd at 1460. The #1 model there is claude-fable-5 at 1507. https://danubelabs.net/en/blog/chinese-ai-models-vs-us-frontier - [2026-09 snapshot] BenchLM's month-end Arena history lists Qwen3.8-Max (Alibaba, proprietary) at 1480, ahead of Kimi K3-max at 1476 and GLM-5.3-max at 1475. That is a different snapshot from the one above, and it means a proprietary Alibaba model can sit above or near Kimi. BenchLM's page text is unclear on which subset this is. In the same snapshot, claude-opus-5-max leads at 1505. https://benchlm.ai/llm-leaderboard-history - [2026-08] A secondary site puts Qwen 3.7 Max at roughly 1488, 9th on the text board. Treat this as low reliability. https://localaimaster.com/blog/lmarena-chatbot-arena-leaderboard - [2026-10-02] Manifold's "#1 company on Arena text on Oct 31" market gives the Oct 2 board as follows. Gemini 4 Argon High (Google) is #1 at 1525. Five Anthropic models follow at 1501–1505. The best OpenAI model is gpt-5.6-sol-xhigh at #20 (1484). So the top of the board is crowded, and Chinese models sit roughly 1475–1490, well below #1. https://manifold.markets/Terminator2/which-company-will-have-the-1-model - [2026-07-16] Moonshot released Kimi K3 (2.8T parameters). It took #1 on Arena's Frontend Code/WebDev board at 1679. Weights were released on Jul 27. That strength is on the coding board, not the text board. https://www.tomshardware.com/tech-industry/artificial-intelligence/moonshot-releases-2-8-trillion-parameter-kimi-k3 - [2026-10-12] A secondary article says Kimi K3 sits at about 1500 on the general text board and has DeepSeek V4 Pro at about 1462. Its date and figures are not consistent with the other sources. Low confidence. https://shattered.io/kimi-k3-coding-leaderboard-2026/ - [2026-10-15] Arena's homepage lists "new release" sessions for MiMo V2.6 Pro (Xiaomi) and DeepSeek V4.1 Flash (Max). It also lists a Gemini 4 Argon first look. Chinese challengers to Kimi were newly added or churning recently. https://arena.ai/leaderboard - [2026-10-15] BenchLM's confirmed-release feed for Oct 9–15 shows no new Moonshot, Alibaba, Z.ai or DeepSeek flagship release. Its Chinese-model ranking, which is not Arena-based, lists MiMo-V2.6-Pro first, then Kimi K3, then Qwen3.8 Max. https://benchlm.ai/best/chinese-models BOTTOM LINE: Kimi K3 is the top Chinese model on Arena's default text view, at about 1488. That lead is narrow, and I could not check it on the style-control-removed view the contract uses. MiMo V2.6 Pro RL (1480), GLM 5.3 (1478) and proprietary Qwen3.8 Max (about 1480) are within roughly 10 points, which is about the size of the board's confidence interval. A new Chinese release before Oct 31 could also change the order. NOT FOUND: The live arena.ai text leaderboard with "Remove Style Control" applied, and its exact current Chinese ranking. Current scores for proprietary Chinese models (Qwen3.8 Max, ByteDance Seed/Dola, Baidu ERNIE) on the default view. Any announced Chinese flagship release scheduled before Oct 31.

Reference class: which Chinese lab holds the top Chinese spot on Arena's text board at a given date. It is a leaderboard-leadership question, so the top spot has been unstable across labs. I could not load the live board (arena.ai returned a Cloudflare challenge), so every score below comes from secondary snapshots. Several of them disagree with each other, and none says whether it uses style control. - [2026-10-04] A snapshot of LMArena's text board stamped 2026-09-02 had claude-fable-5 first at 1507. Kimi K3 was the top Chinese model at 12th (1489, ±5), GLM-5.3 was 20th (1482) and DeepSeek-V4-Pro-0813 was 52nd (1460). The article says Qwen3.8 Max is listed on Arena as proprietary. Source: https://danubelabs.net/en/blog/chinese-ai-models-vs-us-frontier - [2026-09-13] Per an Arena text-leaderboard snapshot cited by aivancity, Qwen3.8-Max (Alibaba) was 22nd at 1481 ±6, with 16,670 votes. That is below the Kimi K3 reading of 12th at 1489 on 2026-09-02. The same article says its launch rank of 5th "was not stable over time". Source: https://aivancity.ai/en/blog/qwen3-8-max-codage-agents-ia-performances-limites/ - [2026-08-03] On launch day Qwen3.8-Max debuted at #5 on the Arena text board with 1496 (preliminary), and was described as the top Chinese model on text. The same Arena post gives the Frontend Code results. Sources: https://felloai.com/qwen-3-8/ and https://x.com/arena/status/2084108703729615026 - [2026-07-16] Kimi K3 launched at #9 on the Arena text board with 1486, up from Kimi-K2.6 at #38. It also took #1 on Frontend Code Arena at 1679. Source: https://x.com/arena/status/2077856214684455116 - [2026-10-04] The text-board lead among Chinese models has changed hands within weeks. Qwen3.8-Max was top Chinese at launch (Aug 3), Kimi K3 reached #9 (Jul 16) and #12 (Sep 2), and GLM-5.3 was about 20th (Sep 2). Moonshot's and Alibaba's scores sit within about 10 points of each other, which is inside the error bars. Source: https://danubelabs.net/en/blog/chinese-ai-models-vs-us-frontier - [2026-04-07] Earlier in the year the top Chinese models on the text board were GLM-5.1 (Zhipu, #14, 1467) and Qwen3.5-max-preview (Alibaba, #15, 1466). Kimi-K2.5-thinking was #25 (1452). Moonshot was therefore not the top Chinese lab then. Source: https://inferencehub.org/blog/chinese-frontier-open-source-ai-models-2026/ - [2025-09 to 2026-04] Earlier leadership was spread across labs: Alibaba's Qwen3-max-preview debuted as top Chinese at #6 in Sep 2025, GLM-5 reached #11 globally by Feb 2026, and GLM-5.1 led in Apr 2026. In each case a launch-week spike was followed by a drop as votes accumulated. Sources: https://www.scmp.com/tech/big-tech/article/3324887/alibabas-qwen3-and-moonshots-kimi-k2-crack-top-10-ai-rankings-closing-us-models and https://inferencehub.org/blog/chinese-frontier-open-source-ai-models-2026/ - [2026-10-12 to 2026-10-16] One source (shattered.io, low reliability: it names GPT-6.1 Sol and conflicts with July reporting) says Kimi K3 is at about 1500 on text. It gives DeepSeek V4 Pro at about 1462 and Qwen 3.7 Max at about 1455 in an "Oct. snapshot". Source: https://shattered.io/kimi-k3-coding-leaderboard-2026/ - [2026-10-04] A separate benchlm.ai history page lists qwen3.8-max at #12 with Elo 1480 and 13,136 votes. The date is unclear, but it is from the last ~6 weeks. It shows Qwen3.8-Max and Kimi K3 within a few points of each other. Source: https://benchlm.ai/llm-leaderboard-history - [2026-09-02] Alibaba's Qwen3.8-Max-0902 debuted #1 on Arena WebDev at 1691, above Kimi K3 (Max). I found no text-board score for it. It is an upgrade that could take the top Chinese text spot, and Alibaba, Zhipu and others ship updates every few weeks. Source: https://x.com/arena/status/2094974637704913198 - [2026-10-11] The Manifold market "SOAI#8" (a Chinese lab overtakes the US frontier on a major leaderboard) was at about 93%. It resolves on the State of AI Report's own grading, which cites Kimi K3's WebDev #1. It is not a text-board market and says nothing about which Chinese lab leads the text board. Source: https://tech-insider.org/china-lab-overtakes-us-ai-leaderboard-manifold-odds/ - [2026-10-04] Artificial Analysis ranks Chinese labs in a different order: MiMo-V2.6-Pro (Xiaomi) 46, GLM-5.3 45, Qwen3.8 Max 45, Kimi K3 44. Xiaomi's MiMo-V2.6-Pro, which was in the top 20 on Arena in September, is another possible challenger. Source: https://danubelabs.net/en/blog/chinese-ai-models-vs-us-frontier BOTTOM LINE: The top Chinese text-board spot is a contested race among Moonshot (Kimi K3), Alibaba (Qwen3.8-Max), Zhipu and others. The leader changed several times in 2026, and the September scores of the leading entries were within about 10 points of each other. Moonshot looks like a plausible front-runner but not a clear one, and a new Chinese release before Oct 31 could displace it. Whether the board is read with or without style control is unconfirmed and could change the order. NOT FOUND: A live Arena text leaderboard reading, with or without style control, for October 2026. Also not found: any Kalshi or Polymarket history for this contract family, and any text-board score for Qwen3.8-Max-0902 or for MiMo-V2.6-Pro.

YES means: On Oct 31, 2026 at 10:00am ET, a Moonshot model (Kimi K3 or a successor) is the highest-ranked model from a mainland-China company on Arena's text leaderboard with "Remove Style Control" applied. - [2026-10 (data through Oct)] The llmrun.dev mirror of Arena text scores shows Kimi K3 at 1488.1, ahead of every other Chinese open model on the board. The next three are MiMo V2.6 Pro RL (1480.4), GLM 5.3 (1478.4) and GLM 5.2 (1475.2). That is a lead of about 8 points over the nearest open rival. The page's open-model list excludes proprietary models such as Qwen3.8 Max. The page states its scores use Arena's default style-control view. https://llmrun.dev/benchmark/lmarena-text - [2026-08-21] Qwen3.8-Max is Moonshot's main proprietary rival. Its launch score of about 1496 was preliminary. It then fell to 1481 ±7 on 9,955 votes by the Aug 21 snapshot (helloai, citing arena.ai), and a Sept 13 snapshot has it at 1481 ±6. That is below Kimi K3's 1488.1 in the Oct mirror. This is the main YES argument: early Qwen scores deflate as votes accumulate. Both figures are with style control on. https://helloai.com/articles/qwen3-8-max-cools-below-gemini - [2026-10-12 (read via search snippet)] benchmarklist.com shows Qwen3.8 Max at 1481.01, with a 95% interval of 1474.2–1487.8 and 9,955 votes. The top of that interval sits just below Kimi K3's 1488 point estimate. The vote count matches the Aug 21 snapshot, so the Qwen score appears to have stopped moving. I could not open the page (HTTP 403). https://benchmarklist.com/models/qwen-qwen3.8-max/ - [2026-10 (BenchLM, month-end history)] BenchLM draws on Arena's published Hugging Face data. It has Qwen3.8-Max at 1480 on 13,136 votes and Kimi K3-max at 1476 on 17,895 votes. Qwen has the higher point score there, but the gap is 4 points, and Kimi has the larger vote base. I could not tell which subset (text or style-control) that row comes from. This is the counter-evidence a YES case has to accept. https://benchlm.ai/llm-leaderboard-history - [2026-07-16] Arena's own post had Kimi K3 launching at #9 with 1486, up from #38 for K2.6. It also had top-10 results in Creative Writing, Coding and Instruction Following. The 1486 launch score is within 2 points of today's 1488, so Kimi's rating has held steady as votes accumulated. https://x.com/arena/status/2077856218262208732 - [2026-07-18] The New Stack reported K3 landed above Claude Opus 4.8 and roughly level with GPT-5.6 Sol on the text board. https://thenewstack.io/kimi-k3-open-weight-coding/ - [2026-10 (Arena changelog)] Arena's changelog lists "kimi-k3 has been added to Text Arena & Code Arena leaderboards" and a later rename to "Kimi K3 (Max)" on the Agent board. That confirms Kimi is a standing entry on the text board, not a stale preview. The changelog is undated here, but it appeared to be several weeks old. https://arena.ai/blog/leaderboard-changelog - [2026-10-16] Raw (no-style-control) ratings are published separately. Arena's Hugging Face dataset has a "text" subset for raw ratings and a "text_style_control" subset. Arena made style control the default on May 16, 2025, with an offset that keeps both views on the same scale. The two views therefore use comparable numbers, and the Remove Style Control view is mostly a small reshuffle, not a different ranking. Whether it favours Kimi or Qwen is unverified. https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset - [2026-09-14] Moonshot shipped Kimi K2.8 Preview on Sept 11. It is a proprietary multimodal model that sits between K2.7 Code and K3. A further Moonshot release before Oct 31 is a possible upside, but I found nothing announced. https://emergent.sh/news/moonshot-ai-launches-kimi-k2-8-preview - [2026-10-16] Single most likely YES path: no Chinese flagship (Qwen3.8-Max-0902 text, MiMo V2.6 Pro RL, GLM 5.3 or a DeepSeek update) gets a text score of about 1488 or more by Oct 31. The Oct 9–15 release feed had none, and Kimi K3's rating has been flat since launch. The Remove Style Control view must also keep Kimi at or above Qwen3.8-Max. Challengers are 8 points or more behind on the default view, and each has a ±5–7 interval. BOTTOM LINE: The strongest YES case is that Kimi K3 is the highest Chinese entry in the default Arena view, at 1488 against about 1481 for Qwen3.8-Max and 1480 for MiMo. Its rating has been flat since launch while Qwen's launch spike deflated. YES then needs only that the order survives the no-style-control view and no rival ships a stronger text model in the next two weeks. NOT FOUND: The live arena.ai text board with "Remove Style Control" applied. A text score for Qwen3.8-Max-0902. Any Oct 2026 text score for ByteDance or Baidu models. Any announced Chinese flagship release before Oct 31. The Oct 12 shattered.io claim that Kimi K3 is at about 1500 on text, which conflicts with other sources and which I do not rely on.

YES means: Moonshot (Kimi) is the highest-ranked mainland-China-headquartered company's model on Arena's text leaderboard with "Remove Style Control" applied, at 10:00 ET on 2026-10-31. - [2026-10-15] The Orcarouter/@teortaxesTex report says DeepSeek may be running an unlabelled new flagship, possibly "V4.1 Pro". It is one observer's X post and is unconfirmed. DeepSeek's changelog has no October entry. The same article confirms V4.1 Pro was named once, in the Sept 10 release note, and has never shipped. A V4.1 Pro could appear on Arena before Oct 31 and take the top Chinese spot. Arena has historically added Chinese flagships at launch, and DeepSeek has tested prototypes there. https://www.orcarouter.ai/blog/deepseek-v4-1-pro-stealth-leak - [2026-10-16] Orcarouter's release list shows Alibaba shipped Qwen3.8 Max (0902) on Sept 2 and Tencent shipped a Hy4 preview on Aug 28. The Sept 2 Arena post says the 0902 build beat Kimi K3 by 17 points on WebDev and beat the earlier Qwen3.8-Max by 22. I found no text-board score for it. Alibaba's launch-day text score of 1496 (Aug 3) was above Kimi K3's 1486 at launch (Jul 16). Alibaba is therefore the main proprietary rival, and its post-training upgrade is the likeliest route to a No. https://x.com/arena/status/2094974637704913198 - [2026-08-03] Arena's own post had Qwen3.8-Max at #5 on text with 1,496, ahead of Kimi K3's July debut of #9 at 1,486. On launch score Moonshot was already the second-best Chinese text entry. The later slide of Qwen to the 1480–1481 range could reverse if the 0902 build settles higher. https://x.com/arena/status/2084108703729615026 - [2026-07-16] Arena's Kimi K3 post stresses Frontend/WebDev strength and category results. Reviewers noted that K3's text score of about 1,486 was "competitive but not dominant" and in the margin of error with GPT-5.6 Sol. Its dominance is on the coding board, not the board this contract uses. https://wan27.org/blog/kimi-k3-benchmarks - [2026-10-15] BenchLM draws on Arena's Hugging Face dataset, which covers both the text and text_style_control subsets. Its Sept snapshot has kimi-k3-max at 1476, called the top open-licence model. The page says openly licensed models are tracked separately, so a proprietary Chinese model such as Qwen can lead without breaking that statement. Two dated snapshots (Sept 2 and Sept 13) put Qwen3.8-Max within about 8 points of Kimi. "Remove Style Control" is a different view from the default, and I could not confirm Kimi's rank on it. https://benchlm.ai/llm-leaderboard-history - [2026-10-15] Arena's homepage shows live sessions for MiMo V2.6 Pro (Xiaomi) and DeepSeek V4.1 Flash. Its "Leaderboard Trends" lists Kimi K3, GLM 5.2 and DeepSeek V4 Flash as the Chinese entries. Vals.ai dates MiMo V2.6 Pro to Sept 18 and V4.1 Flash to Sept 10. Chinese challengers have been shipping about every 2–3 weeks, so a further release before Oct 31 is likely. Kimi's lead depends on none of them beating it. https://arena.ai/leaderboard - [2026-09-10] DeepSeek's changelog shows releases on Jul 31, Aug 13, Aug 21 and Sept 10, which is a 20-day gap between the last two. That cadence makes another DeepSeek update before Oct 31 plausible. https://api-docs.deepseek.com/updates/ - [2026-10-12] Shattered.io says K3 scores about 1,500 on text, but the same article calls the Frontend Code 1,679 result "Oct 12" news when it is a July result. It is therefore unreliable as support for Kimi's current standing. A stale Kimi lead claim is not evidence for Yes. https://shattered.io/kimi-k3-coding-leaderboard-2026/ - [2026-10-02] Manifold's market on the text #1 company is a sibling and shows Chinese labs as an afterthought. It puts "Other" at 1.8% for #1 and confirms that the Oct 2 board has the Chinese models far below the top. It does not decide which Chinese lab is first. The contract's "highest ranked Chinese" depends on a cluster of 1475–1490 scores, where a 3–5 point shift flips the winner. https://manifold.markets/Terminator2/which-company-will-have-the-1-model - [2026-10-16] Technicalities that favour No: - The resolution time is a single instant (Oct 31, 10:00 ET). - "Remove Style Control" is a non-default view where ranks differ. - Tie rules mean a shared rank would pay Moonshot only a fraction, not full Yes. - Arena rank is computed with confidence intervals, so rank-sharing among Chinese entries is possible. - If Qwen3.8-Max or MiMo is tied or ahead, that is not Yes. Style-controlled boards usually penalise verbose, formatted models. I found no evidence of how K3 fares relative to Qwen on the non-style-controlled view. https://benchlm.ai/llm-leaderboard-history BOTTOM LINE: Moonshot leads the default text board only narrowly. Qwen3.8-Max (0902 build), MiMo V2.6 Pro, GLM 5.3 and a possible DeepSeek V4.1 Pro all sit within a few points, and a Chinese release before Oct 31 could displace it. Kimi's rank on the "Remove Style Control" view is unverified, and tie rules reduce a payout. NOT FOUND: The live arena.ai text board in the "Remove Style Control" view. A text-board score for Qwen3.8-Max-0902 or MiMo V2.6 Pro. Any confirmed Chinese flagship release date before Oct 31. Any Kalshi history for this contract.