forecasting
agent
← all forecasts

Will no company have an AI model hit 1550 on Chatbot Arena in 2026?

polymarketCompanies2026-11-13llm cost $1.90view on polymarket ↗
52%
Agent
36%
Market Price
+17.0pp
Edge
42%
Confidence
31,925
Volume
23.0c
Spread
48
Days to resolution
32
Markets in event
final rationale
average of 2 draws (52%, 53%), redrawn after an extra research pass
already resolved?
No source shows a 1550+ score on the overall no-style-control board; LMArena's HF dataset (2026-11-11) has Gemini 4 Carbon at 1538.38 (CI to 1547.9, ~4.3k votes, preliminary), so the market is unresolved.
time left
About seven weeks remain, to Dec 31, 2026.
if nothing changes
Status quo is 'None in 2026', with the leader 12 points short.
path to no
Carbon's preliminary score keeps rising or swings briefly to 1550 (a transient print counts), Fable 5.2 debuts high, or a December release (Opus 5.6, a new Gemini variant, GPT-6.1) beats Carbon by 12+ points, larger than the usual 4–9 point leader steps.
path to yes
Carbon reverts down like Argon did, Fable 5.2 lands near Anthropic's usual 1512–1514, and any December model scores lower or lacks settled votes by Dec 31. The market mid is 35.5% on a very wide 24/47 spread (last trade 43%), while the sibling company markets sum to about 44% for some company getting there, implying about 56% for None. My estimate (about 15% Carbon drift plus 20–30% from new releases) puts Yes slightly above even.
price vs. prediction
market price agent forecast at t+0
8 follow-ups
earlier forecasts on this market
datepredictedmarket priceconfidence
2026-10-29 70% 66% 56%
2026-08-27 84% 85% 65%
2026-08-16 81% 80% 56%
1. Planner Opus 5.5
Tools chosen: polymarket_direct resolution_source polymarket_related kalshi_related article_search
Planner reasoning
The market resolves on the LMArena text leaderboard with style control off, so the current top score shown there is the key fact. Polymarket's own price is the anchor, and related markets (top AI model at year end, other score thresholds) add cross-market signal. One targeted search for recent top-model Arena scores can confirm how close the leader is to 1550.
2. Research 4 web research agents, Sonnet 5.5
summary
RESEARCH DIGEST (built in code: tool summaries and each report's bottom line; the forecasters read the full tool outputs and reports) Tools - polymarket_direct: ## This Market's Polymarket Data **Will no company have an AI model hit 1550 on Chatbot Arena in 2026?** - Current price (probability): 35.50% (live mid; bid 24% / ask 47%; last trade 43%) - 7-day price change: -15.50% - 30-day price change: -40.50% - Total volume: $31,925 (USD notional) - Price range: 34.50% - 87.50% - Data points: 84 days - resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=2026-11-11. LMArena's official Hugging Face leaderboard dataset (config "text", split "latest", category "overall", the non-style-control counterpart of the separate "text_style_control" config), published 2026-11-11, shows Gemini 4 Carbon (High) from Google at #1 with 1538.38 (CI 1528.82–1547.93, 4,273 votes). Gemini 4 Argon (High) is - polymarket_related: Polymarket search over 4 keyword(s), kept 17 matches. keyword 'Chatbot Arena': 5 markets | keyword '1550': 5 markets | keyword 'best AI model end of': 5 markets | keyword 'top AI model': 2 markets - kalshi_related: 2 related markets / summaries. keyword 'LMArena': no matches | keyword 'top AI model': ok | keyword 'Chatbot Arena': ok - article_search: Fetched 16 unique articles across 2 queries (lookback=30d). 'LMArena text leaderboard top Arena score November 2026 without style control': 8 articles | 'new model release Arena score Gemini GPT Claude Grok': 8 articles Research loops, bottom lines - current state: As of Nov 13 the best overall text score is about 1538 (Gemini 4 Carbon, preliminary, ±10), roughly 12 points short of 1550, with seven weeks left. Carbon's score is still moving as votes arrive, and newer Anthropic models (Fable 5.2, Mythos 5.2) reportedly released in November have no Arena score yet. The tracker data may not exactly match the no-style-control view the market resolves on. - base rate: The top score is 1526–1538, roughly 12–24 points below 1550, with about 7 weeks left. History shows ~11 points a month on average but a plateau of about 1500–1538 since Feb 2026. Whether 1550 gets hit depends on one more strong release or a score drift on Carbon or another top model. - yes case: The best evidence for "None in 2026" is that the top overall score is a preliminary 1538 on about 4.3k votes with no model within 20 points of it from another lab. Preliminary scores tend to drift down as fresh votes arrive, as Argon's did. A stable 1550 would need a 12-point jump in 7 weeks, against a recent pace of about 4 points a month. The main risk to this case is Fable 5.2's unscored text debut; the one thin contract on that strike prices 1550 near 2%. - no case: The leader sits about 12 points below 1550 on a preliminary score with a wide interval. The changelog shows no new Text Arena frontier entry since Carbon, and the calendar's next Anthropic flagship, an unconfirmed Dec 21 estimate, leaves little time. Status quo and regression of preliminary scores both favour NO, though a Fable 5.2 text listing or a Carbon upswing remain the credible routes to YES's opposite. Crux follow-up - The status quo favours "None in 2026". The leader is 12 points below 1550, its confidence interval tops out near 1548, and the next-best non-Google models are at about 1512–1514. A Yes needs Carbon to drift up, a higher Gemini variant, or a December flagship that beats Carbon by a wide margin, and nothing confirmed points to that. The risk is real but not strong. Fable 5.2's text debut is unknown, and a single transient 1550 would count.
tools
toolstatustimeitemssummary
polymarket_direct OK 0.7s 1 ## This Market's Polymarket Data **Will no company have an AI model hit 1550 on Chatbot Arena in 2026?** - Current price (probability): 35.50% (live mid; bid 24% / ask 47%; last trade 43%) - 7-day price change: -15.50% - 30-day price change: -40.50% - Total volume: $31,925 (USD notional) - Price ra
resolution_source OK 73.7s 1 already_resolved=unclear; matches_criteria=True; last_updated=2026-11-11. LMArena's official Hugging Face leaderboard dataset (config "text", split "latest", category "overall", the non-style-control counterpart of the separate "text_style_control" config), published 2026-11-11, shows Gemini 4 Carbo
polymarket_related OK 0.6s 17 Polymarket search over 4 keyword(s), kept 17 matches. keyword 'Chatbot Arena': 5 markets | keyword '1550': 5 markets | keyword 'best AI model end of': 5 markets | keyword 'top AI model': 2 markets
kalshi_related OK 5.1s 2 2 related markets / summaries. keyword 'LMArena': no matches | keyword 'top AI model': ok | keyword 'Chatbot Arena': ok
article_search OK 0.4s 16 Fetched 16 unique articles across 2 queries (lookback=30d). 'LMArena text leaderboard top Arena score November 2026 without style control': 8 articles | 'new model release Arena score Gemini GPT Claude Grok': 8 articles
web research agents 4/4 ok, 16 searches, 2 page reads
dimensionstatustimesearchespage readsreport
current state OK 21s 3 0 4085 chars
base rate OK 28s 4 1 5235 chars
yes case OK 35s 5 1 6144 chars
no case OK 23s 4 0 5540 chars
report: current state
- [2026-11-13] The top of the Text Arena overall board is Google's Gemini 4 Carbon (High) at 1538±9.6 (4,273 votes). Gemini 4 Argon is second at 1529. Claude Opus 5.5 is third at 1514. The aggregator reads this from LMArena's Hugging Face dataset, so it is a secondary source. Nothing is at 1550 yet. https://llmbenchmark.ai/benchmarks/lmarena-text - [2026-11-11] The arena.ai snapshot of the overall text board, with 9.12M votes and 421 models, shows the same order. Carbon is on 1538±10, Argon on 1529±5, Opus 5.5 on 1514, Fable 5.1 on 1512, Opus 5 on 1507, Meta Muse Wonder on 1503 and Gemini 4 Hydrogen on 1502. The URL carries `scoring=raw`, which may be a different view from the resolution page. https://arena.ai/leaderboard/chat/text?scoring=raw - [2026-11-06/07] Carbon was 1526±9 and tagged "Preliminary". Argon was 1523 and Meta's Muse Wonder 1513. So Carbon's score rose about 12 points in a week as votes accumulated. The source is a third-party tracker citing the official board. https://www.developersdigest.tech/benchmarks/lmarena - [2026-10-27/28] Carbon was at 1528 and Argon at 1520. https://chatjudge.ai/benchmarks/chat/lmarena-text - [2026-10-20] A tracker reported Argon at 1532 on the board with style control off, which is the resolution view. It noted that Argon was scored on Arena from Sept 30, before its Oct 13 public release, so the score was preliminary. If accurate, that means a no-style-control score near 1532 had already appeared; it is a third-party report. https://www.versusapps.com/blog/lmarena-leaderboard-september-2026 - [2026-11-08] The arena.ai lab-rank view put Google first with Carbon at 1527±10. OpenAI's best was GPT-5.6 Sol at 1485 and Alibaba's Qwen3.8-Max at 1484. Neither is near 1550. https://arena.ai/leaderboard/chat/text?rankBy=labs - [2026-11-07] In the "expert" subcategory, Argon scored 1554 and Opus 5 scored 1553. This is not the overall board and does not count for resolution. https://llmbenchmark.ai/benchmarks/lmarena-text-expert - [2026-11-11] On the instruction-following subcategory, Argon led at 1528 and Carbon was at 1525. The new Gemini 4 Hydrogen is in the top 10 there. This is also not the resolution board. https://www.benchleader.com/benchmarks/lmarena_text_if - [2026-11] A less reliable aggregator (swfte) lists Claude Fable 5.2 and Mythos 5.2 as November 2026 releases with no Arena score yet. The same page claims Claude Mythos 5 is at 1531, which conflicts with the official-board trackers above, so treat it as unverified. Fable 5.1 (Max) is on the board at 1512. https://www.swfte.com/lmarena - [2026-02 to 2026-08] Scores have crept up only slowly this year. Gemini 3.1 Pro and Opus 4.6 were at about 1500–1504 in February. Claude Fable 5 was at about 1525 in August after a July 12 re-baseline. The Arena "Max" router hit 1500 in about August. The top is now about 1538, so the ceiling has risen roughly 35–40 points in 9 months. https://x.com/arena/status/2024519891295089063 - [2026-11-13] I could not open arena.ai directly because it returned 403 to fetch. All current board figures come from trackers that cite it. The Hugging Face dataset used by trackers may show the default style-control view rather than the no-style-control one the market uses. The market uses the no-style-control view, and Argon's reported 1532 there on Oct 20 suggests the two views are close. https://arena.ai/leaderboard/text/overall-no-style-control BOTTOM LINE: As of Nov 13 the best overall text score is about 1538 (Gemini 4 Carbon, preliminary, ±10), roughly 12 points short of 1550, with seven weeks left. Carbon's score is still moving as votes arrive, and newer Anthropic models (Fable 5.2, Mythos 5.2) reportedly released in November have no Arena score yet. The tracker data may not exactly match the no-style-control view the market resolves on. NOT FOUND: Direct reading of the arena.ai no-style-control page (403). Official confirmation of Fable 5.2 or Mythos 5.2 Arena scores. Any announced model launch dates between now and Dec 31. Any report of a score of 1550 or more on the overall text board.
report: base rate
- [2026-11-11] Arena's own Text Arena overall page, default view, shows Gemini 4 Carbon (High) first at 1526±10. Gemini 4 Argon (High) is at 1522 and Meta's Muse Wonder (Max) at 1511. The ±10 interval and the low vote count (4,273) suggest the score is still preliminary. A second snapshot of the same page, the "?scoring=raw" URL, shows the top at 1538±10, with Argon at 1529 and Claude Opus 5.5 at 1514. I could not tell which view is style control on or off. The resolution page, overall-no-style-control, returned a Cloudflare block, so I could not read it directly. https://arena.ai/leaderboard/chat/text - [2026-11-12] BenchLM tracks Arena data fetched 2026-11-12 and lists Gemini 4 Carbon (High) at 1538 and Argon at 1530. No model is listed at 1550 or above. https://benchlm.ai/llm-leaderboard-history - [2026-11-12] BenchLM's milestone table gives the first month the #1 model crossed each threshold. The first 1500 was claude-opus-4-6 in 2026-02. The first 1450 was gemini-2.5-pro in 2025-06. The first 1400 was in 2025-03. Those dates imply about 8 months from 1450 to 1500 and about 9 months from 1500 to a 1538 top. https://benchlm.ai/llm-leaderboard-history - [2026-11-12] Over the same history, the #1 score rose from 1075 to 1538 in 43 months (+463), which BenchLM calls roughly 11 points a month on average. The leader changed 21 times among 7 providers, and the longest reign was 5 months. At the average rate, the 12–24 points still needed would take about 1–2 months. However, top scores have mostly sat between 1500 and 1538 since Feb 2026, so the recent rate looks slower. https://benchlm.ai/llm-leaderboard-history - [2026-02-20] The 1500 threshold was first reached in early 2026. When Gemini 3.1 Pro launched, Arena said it was tied #1 in Text at 1500, 4 points behind Opus 4.6. https://x.com/arena/status/2024519891295089063?lang=en - [2025-11 / 2026-11] Gemini 3 Pro launched at a reported 1501, which was a headline "breakthrough" figure at the time. Eleven months later the top is about 1526–1538. That is roughly +25 to +37 points over 12 months for the very top model. https://blog.google/products/gemini/gemini-3/ - [2026-10-20 / 2026-10-28] Top scores in October moved around: Argon was 1532 with style control off on Oct 20, and Carbon was 1528 on Oct 28. Third-party trackers show the top staying at 1526–1532 for about 3 weeks, which points to a plateau. Argon's early score was labelled preliminary. Argon was scored from Sept 30 and released publicly Oct 13. https://www.versusapps.com/blog/lmarena-leaderboard-september-2026 https://www.benchleader.com/benchmarks/lmarena_text - [2026-11-12] Gemini 4 Hydrogen was released on Nov 12 and is also on Arena. A snapshot of the leaderboard from Nov 11 shows Hydrogen (High) at 1502, so it is not a threat. Gemini 4 Carbon has only about 4.3k votes. Its score can still move, and so can the other Gemini 4 and GPT-6 variants. https://arena.ai/leaderboard/chat/text?scoring=raw - [2026-11-11] Specialist boards show higher numbers, but they don't count here. GPT-6 Astra (Max) is at 1548 on Text Arena Coding. Text Arena Expert shows a 1517 top within its snippet. The Arena Expert board put Opus 4.6 near 1538 back in Feb. Only the Overall no-style-control board counts. https://arena.ai/leaderboard/chat/text/coding - [2026-11-13] Polymarket's sibling market on which company has the best AI model at end of 2026 priced Google at 75%, Anthropic at 19% and OpenAI at 3%, per a 2026-11-13 aggregator snapshot. A 2026-10-01 snapshot had Anthropic at 53% and Google at 39.5%, so the favourite flipped to Google after Gemini 4 launched. That market resolves on rank, not score, so it says little about a 1550 threshold. https://coinstats.app/prediction-markets/odds/tech/which-company-has-best-ai-model-end-of-2026/ - [2026-10-20] A related prediction page (Versus Apps, an unverified aggregator) lists "Arena score" answer strikes of 1520, 1530, 1540, 1550, 1600, 1650 and 1700. Its strike ladder suggests 1550 is about 1.5% to 3% above the top, in the tail of what was expected. I did not find that market's prices. https://www.versusapps.com/blog/lmarena-leaderboard-september-2026 - [2026-11-04] Some aggregator pages disagree on the top score. One lists a "Claude Mythos 5" at 1531, and WebDev Arena shows Claude Opus 5.5 (Max) at 1810, but that board is a different arena. Aggregator data is noisy and sometimes inconsistent with Arena's own page, so treat third-party numbers with care. https://www.swfte.com/lmarena BOTTOM LINE: The top score is 1526–1538, roughly 12–24 points below 1550, with about 7 weeks left. History shows ~11 points a month on average but a plateau of about 1500–1538 since Feb 2026. Whether 1550 gets hit depends on one more strong release or a score drift on Carbon or another top model. NOT FOUND: I could not read the no-style-control resolution page, because it was blocked. I could not tell whether style control was on or off in the snapshots, and the 1526 versus 1538 gap for Carbon may reflect exactly that. I found no Polymarket pricing for the 1550-threshold market itself (deliberately not searched). I found no confirmed upcoming frontier releases, such as a GPT-6.x or Claude release, dated for Nov–Dec 2026.
report: yes case
YES means: No model reaches an Arena Score of 1550 or more on the Text Arena overall board with style control off by Dec 31, 2026 11:59 PM ET, so the market resolves "None in 2026". - [2026-11-12] BenchLM's full table, fetched, puts the top five at 1538 (Gemini 4 Carbon), 1530, 1513, 1512 and 1506. Nothing from Anthropic, OpenAI, Meta or the open-weight labs is within 25 points of 1550. The one model within 12 points is Carbon, which has only about 4,282 votes. Fixing the "preliminary" reading needs a drop of just a few points. Carbon is not a stable 1538 reading. https://benchlm.ai/llm-leaderboard-history - [2026-11-12] BenchLM's milestone table shows a slow ladder: 1450 in 2025-06, 1500 in 2026-02, and no 1550 yet. The #1 score has moved about 38 points in 9 months, far below the 11/month long-run figure the colleagues cited. 1550 needs about +12 in 7 weeks, against a recent pace of roughly 4 per month. https://benchlm.ai/llm-leaderboard-history - [2026-11-13] The only new Anthropic model, Claude Fable 5.2, launched Nov 9 and the Nov 9 Arena snapshot did not rank it for text. Arena's own front page already lists it #1 in WebDev and Agent Arena, so the text score is probably imminent. The predecessor Fable 5.1 sits at 1512, and Opus 5.5 at 1513–1514 is Anthropic's best text score. Fable 5.2 would need about +38 over Fable 5.1 to reach 1550. https://arena.ai/leaderboard and https://projetcelsius.com/en/blog/claude-fable-5-2/ - [2026-11-13] A Public.com prediction contract on the Fable 5.2+ debut score (no style control, before Jan 1, 2027) priced "Above 1550" at about 2¢. The same page priced "Above 1540" at 95¢, which looks inconsistent. Total volume was only $166, so it is thin and unreliable. It is the only market I found aimed at this exact 1550 strike. https://public.com/prediction-markets/fable-5-2-debut-arena-score-sqjn2ir?sourceScreen=prediction_markets_event_page - [2026-11-02] Carbon is not generally available. LLM Odds reports that only vetted cyber defenders in Google's Fairwind Program have access, with no wider release date. That limits fresh public votes. The lead rests on about 4,400 votes and is labelled preliminary. Arena's changelog says a preliminary score "may shift" after public launch, and it describes a pre-release testing effect that shrinks as fresh votes arrive. Arena's own rebuttal puts the pre-release testing effect at about +11 Elo after 50 tests and 3,000 votes (technolynx summary). That is close to the 12-point gap, so mean reversion is a plausible mechanism. https://llmodds.com/news/google-gemini-4-carbon-launch-2026, https://arena.ai/blog/leaderboard-changelog and https://www.technolynx.com/post/what-chatbot-arena-is-and-how-it-works-and-what-it-can-t-tell-you-about/ - [2026-10-20] The same pre-release dynamic already played out once. Argon was 1532 on the no-style-control board on Oct 20, was marked preliminary, then settled to 1529–1530 by Nov 12. The Oct 28 BenchLeader snapshot labels its Text board "with style control" and shows Carbon at 1528. Trackers reading the dataset therefore carry a roughly 1520–1538 spread for the top two models. That spread has not produced a 1550. The Argon figure comes from the unverified Versus Apps page, which I could not fetch (403). https://www.versusapps.com/blog/lmarena-leaderboard-september-2026 and https://www.benchleader.com/benchmarks/lmarena_text - [2026-11-12] Gemini 4 Hydrogen, released Nov 12, sits at 1502 on the Nov 13 llmbenchmark.ai table. Google has two more top models, Carbon and Argon, but its lower tier is not a path to 1550. https://llmbenchmark.ai/benchmarks/lmarena-text - [2026-11-11] The only 1550+ numbers anywhere are in other categories: coding (Carbon 1573, Argon 1550 per BenchLeader, Oct 28) and expert, plus Multi-Turn 1568 and Chinese 1614 on BenchLM. None counts for this market, which resolves only on the Overall no-style-control board. The scores in the English and Hard Prompts categories on BenchLM (1531 and 1546) are also below 1550. Category numbers exceeding 1550 show the scale can stretch, but they do not move the overall board. https://www.benchleader.com/benchmarks/lmarena_text_coding and https://benchlm.ai/llm-leaderboard-history - [2026-11-13] The 2026 release cadence shows the board does not jump with each release. OpenAI shipped GPT-6 Astra on Sep 4 and Sol and Luna on Sep 22. Anthropic shipped Fable 5.1 on Sep 1 and Fable 5.2 on Nov 9. Meta shipped Muse Wonder. After all of these the top overall score is still 1538 at best. OpenAI's best, GPT-6 Sol, was at 1485 on Nov 8 per the earlier report, 65 points short of 1550. The Wikipedia GPT-6 page already lists a GPT-6.1 successor, but I found no scored GPT-6.1 text result. https://en.wikipedia.org/wiki/GPT-6 - [2026-11-13] Only 7 weeks remain, and the usual December flagship window is the only realistic catalyst. I found no dated launch for a new frontier model, and the only reported upside is the unscored Fable 5.2. A 1550 outcome needs a roughly +12 point drift or a fresh model beating the leader by 12 or more. History on this board has seen the leader beat the prior leader by 4–9 points. Examples are Carbon over Argon by 8–9 points and Opus 4.6 over Gemini 3.1 Pro by 4 in February. An increase of 12 or more over the leader would be a larger-than-typical step. https://llmodds.com/news/google-gemini-4-carbon-launch-2026 and https://x.com/arena/status/2024519891295089063 BOTTOM LINE: The best evidence for "None in 2026" is that the top overall score is a preliminary 1538 on about 4.3k votes with no model within 20 points of it from another lab. Preliminary scores tend to drift down as fresh votes arrive, as Argon's did. A stable 1550 would need a 12-point jump in 7 weeks, against a recent pace of about 4 points a month. The main risk to this case is Fable 5.2's unscored text debut; the one thin contract on that strike prices 1550 near 2%. NOT FOUND: Fable 5.2's text Arena score. A direct read of the no-style-control page (arena.ai blocked me and Versus Apps returned 403). A confirmed Nov–Dec release date for any new frontier model. Any overall text score of 1550 or more.
report: no case
YES means: no model reaches an Arena Score of 1550 or more on the Text Arena overall board (style control off) by Dec 31, 2026, so the market resolves "None in 2026". - [2026-11-12] The Arena changelog, which I read directly, shows Claude Fable 5.2 (Max) was added to the Agent Arena on Nov 11 and the WebDev board on Nov 9. It has not been added to the Text Arena. The Text Arena entries since Oct 28 are only Kimi K3.1 Pro (Nov 9), plus Claude Haiku 5.5 and beam-1.0 in October. The only new frontier text candidate has no text listing yet, and Hydrogen (1502) is not a threat. https://arena.ai/company/leaderboard-changelog - [2026-10-28] Carbon was added to the Text Arena on Oct 28 per the changelog. The first read was 1528, and it was 1526±9 on Nov 6. The rise to 1538 (the llmbenchmark.ai figure, 4,273 votes, ±9.6) is therefore not a monotonic trend. Early votes on the Oct 28 listing were counted at 1528, and Carbon is a restricted-access model. Per Arena's policy, scores built on pre-release votes stay preliminary until fresh post-release votes arrive, and these can pull an inflated score down. Carbon's Text Arena votes are about 4.3k–4.5k (Nov 6/13 trackers), so the interval is still wide. https://arena.ai/blog/policy - [2026-10-28] Carbon is restricted to vetted cyber defenders, with no wider release date set. Arena's rule requires public release before a score is final, so its preliminary tag may last. A wider release would also bring in more fresh votes, which tend to pull a preliminary score toward its true mean. https://llmodds.com/news/google-gemini-4-carbon-launch-2026 - [2026-10-20] Style control matters in the resolution view. Per the Oct 20 tracker, Argon read 1532 with style control off and 1524 with it on, a +8 gap. The Nov 13 trackers show Carbon at 1538 and Argon at 1529, and the Nov 11 Arena default view showed 1526 and 1522, so the 1538 figure is probably the no-style-control view. That would put the real gap to 1550 at 12, not smaller. Argon's no-style-control gap over the default view is the only one on record, and I found no evidence that Carbon's gap is larger. https://www.benchleader.com/benchmarks/lmarena_text - [2026-11-05] The Arena scale has compressed. In the 2026-10-28 Hugging Face dataset, Carbon is rank 2 at a normalised score of about 0.16, a different scale from the displayed Elo-style one. Carbon sits only 8–9 points over Argon and about 24 over the best non-Google model (Opus 5.5 at 1514). Gemini 4 Argon was 1524 on Oct 15 and 1529 on Nov 13. That is only about 5 points of drift in a month for a model with 20k votes. A stable, well-voted model drifts slowly. https://codelifter.com/models/ - [2026-11-12] Release calendar. A forecast calendar from WaifuAI (estimates, not announcements; updated Nov 12) lists no Anthropic frontier text model before Claude Opus 5.6 on Dec 21. Its December OpenAI entries are GPT-6.1 Astra and GPT-6.2 Sol on Dec 10, with GPT-7 only on Jan 20, 2027. A model released on Dec 21 has about 10 days for votes to settle before Dec 31. GPT-6 Sol, OpenAI's best, is about 50 points below the top (1485 on Nov 8). Arena's own policy treats ratings as needing at least 1,000 votes to stabilise. https://waifuai.github.io/llm-countdowns/ - [2026-11-05] Specialist boards (Coding 1573, Expert 1554) are higher, but the market names only the Text Arena overall board with style control off, so these do not count. The coding-board spread over overall is large, about 45 points for Carbon (1573 vs 1528 on Oct 28). That shows how far apart categories sit from the overall board and why headline "1550+" mentions may not be on the resolution board. https://www.benchleader.com/benchmarks/lmarena_text_coding - [2026-11-13] Base-rate arithmetic. The top rose about 38 points in 9 months, from 1500 in Feb to 1538 now. Against that, a 12-point gain in 7 weeks needs about 2.5 times the recent pace, roughly 4 points a month. Recent increments came in steps tied to a single new Google model: Argon reached 1524–1532 in October and Carbon 1528 on Oct 28. The pace has been about 8–12 points per release, so a 1550 print most plausibly needs a new frontier release that beats Carbon by 12+. Carbon's own score is the only current path, and a statistical-noise drift of +12 on a ±10 interval is not a repeatable event. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-11-12] Verification risk. The resolution rule says the first entity to reach 1550 on the leaderboard. A transient preliminary value shown on a tracker, as with Argon's pre-release 1532, may not be captured on the resolution page itself. Third-party reports mix default, raw and no-style-control views, so any single claim of 1550 needs checking against the official page. I could not read that page directly. https://arena.ai/blog/leaderboard-changelog BOTTOM LINE: The leader sits about 12 points below 1550 on a preliminary score with a wide interval. The changelog shows no new Text Arena frontier entry since Carbon, and the calendar's next Anthropic flagship, an unconfirmed Dec 21 estimate, leaves little time. Status quo and regression of preliminary scores both favour NO, though a Fable 5.2 text listing or a Carbon upswing remain the credible routes to YES's opposite. NOT FOUND: A direct read of the no-style-control page (403 blocked), and the versusapps page when fetched (403). Any Text Arena score for Fable 5.2 or Mythos 5.2. Any confirmed release date for a frontier model in Nov–Dec 2026. Any report of 1550 or more on the overall text board.
3. Forecasters 2 draws, Opus 5.5 + Fable 5.2, re-drawn after crux research
crux research
Crux research ran because the draws' mean 0.54 sits 0.19 from the same-question market "this polymarket market" at 0.35. One more research agent looked into that point, and every draw below was redone with its report.
first-round draws: 50%, 58% , crux research: 4 searches, 1 page read
report: crux
CRUX: Where Fable 5.2's pending no-style-control text score lands, and whether Gemini 4 Carbon's preliminary 1538 drifts up or down. Carbon is 12 points short, and Fable 5.2 launched on Nov 9 with no official text score yet. Evidence: - **2026-11-13, leader at 1538.** A tracker reading LMArena's Hugging Face dataset shows Gemini 4 Carbon led with 1538, ahead of Gemini 4 Argon at 1529 . Carbon's row is 4,273 votes with a ±9.6 interval, and Opus 5.5 (1514) and Fable 5.1 (1512) follow. Nothing is at 1550. URL: https://llmbenchmark.ai/benchmarks/lmarena-text. I could not open arena.ai directly; it returned 403 and a Cloudflare challenge. I also can't confirm from this source that the 1538 is the no-style-control view. - **Carbon launch, Oct 28.** Arena put Carbon (High) on 1,528 points, just ahead of Gemini 4 Argon on 1,520 . https://llmodds.com/news/google-gemini-4-carbon-launch-2026 - **Carbon's score rose about 10 points in 2 weeks.** A Nov 6 Arena snapshot showed 1526±9 (Preliminary) for Carbon on 4,483 votes, on the default style-control view. That version is not identical to the resolution view. https://www.developersdigest.tech/benchmarks/lmarena. Another tracker saw the style-control version at 1524 in October. https://www.versusapps.com/blog/lmarena-leaderboard-september-2026 - **Carbon's access is limited.** Only vetted cyber defenders can use it, and Google has set no wider release date. A wider release could change who votes on it and how many votes it gets. - **Argon's score drifted both ways.** One tracker had Argon at 1520 on Oct 28. Another had it at 1532 on a snapshot of unknown date, and the Nov 13 dataset has 1529. https://sophon.at/leaderboards/lmarena-text. Preliminary scores can move several points either way. - **Fable 5.2 has no official text score yet.** It was released on 9 November 2026 . Arena announced it #1 in the Code Arena: WebDev with 1832 pts . A Nov 11 snapshot shows it only on the Agent Arena board. https://x.com/arena/status/2119948067449536733, https://arena.ai/leaderboard/agent. Anthropic's recent text scores are 1512–1514 (Fable 5.1, Opus 5.5), even though the newer models do much better on coding and agent benchmarks. - **One unofficial aggregator lists Fable 5.2 at 1538 (Nov 9).** It lists Carbon at 1528 and Opus 5.5 at 1526, which doesn't match the official dataset, so I treat it as unreliable. https://openlm.ai/chatbot-arena/ - **Possible December releases.** Kalshi shows Opus 5.6 at 84% to release before Dec 16, 2026 , with only 1.2K traded. A countdown tracker's estimate of Dec 21 is a model-based guess, and Anthropic has given no date. https://kalshi.com/release-dates, https://waifuai.github.io/llm-countdowns/anthropic/claude-opus-5-6/. A Public.com market on Fable 5.2's debut text score priced "above 1550" at 2%. That market has very thin volume (about $166). https://public.com/prediction-markets/fable-5-2-debut-arena-score-sqjn2ir BOTTOM LINE: The status quo favours "None in 2026". The leader is 12 points below 1550, its confidence interval tops out near 1548, and the next-best non-Google models are at about 1512–1514. A Yes needs Carbon to drift up, a higher Gemini variant, or a December flagship that beats Carbon by a wide margin, and nothing confirmed points to that. The risk is real but not strong. Fable 5.2's text debut is unknown, and a single transient 1550 would count. NOT FOUND: - Fable 5.2's official text score, either style-control setting. - A direct read of the arena.ai no-style-control page. - Any confirmed December release date for Opus 5.6, GPT-6.1 or a new Gemini. - Carbon's vote count and score trend after Nov 11.
claude-opus-5-5, high effort
42% conf
Yes 52%
No 48%
already resolved?
No source shows a 1550+ score on the overall no-style-control board; LMArena's HF dataset (2026-11-11) has Gemini 4 Carbon at 1538.38 (CI to 1547.9, ~4.3k votes, preliminary), so the market is unresolved.
time left
About seven weeks remain, to Dec 31, 2026.
if nothing changes
Status quo is 'None in 2026', with the leader 12 points short.
path to no
Carbon's preliminary score keeps rising or swings briefly to 1550 (a transient print counts), Fable 5.2 debuts high, or a December release (Opus 5.6, a new Gemini variant, GPT-6.1) beats Carbon by 12+ points, larger than the usual 4–9 point leader steps.
path to yes
Carbon reverts down like Argon did, Fable 5.2 lands near Anthropic's usual 1512–1514, and any December model scores lower or lacks settled votes by Dec 31. The market mid is 35.5% on a very wide 24/47 spread (last trade 43%), while the sibling company markets sum to about 44% for some company getting there, implying about 56% for None. My estimate (about 15% Carbon drift plus 20–30% from new releases) puts Yes slightly above even.
claude-fable-5-2, high effort
42% conf
Yes 53%
No 47%
already resolved?
Not resolved. LMArena's own Hugging Face dataset (published 2026-11-11) shows the Text Overall leader, Gemini 4 Carbon (High), at 1538.38 (CI 1528.8-1547.9, 4,273 votes), and no tracker reports any overall score at or above 1550. The official no-style-control page itself could not be read (403).
time left
About seven weeks remain to Dec 31, 2026 11:59 PM ET.
if nothing changes
If nothing changes, the top stays around 1538 and the market resolves Yes ("None in 2026"): the top has risen only about 38 points in nine months (~4 a month), and the next-best non-Google models sit at 1512-1514.
path to no
Any single reading of 1550+ counts, so it could come from Carbon drifting up 12 points on a preliminary, low-vote score, a strong text debut for the still-unscored Claude Fable 5.2, or a December flagship such as Opus 5.6 (Kalshi 84% to release before Dec 16, thin volume), GPT-6.1 or a new Gemini variant beating the leader by 12 or more.
path to yes
Carbon's preliminary score holds or drifts down as fresh votes arrive (Argon went from 1532 to 1529), Fable 5.2 lands near Anthropic's usual 1510-1525 text range, and any late-December release lacks the margin or the time to print 1550. The market mid is 35.5%, but the book is wide (24/47), the last trade was 43%, and volume is thin; it has fallen 40 points in 30 days as Gemini 4 arrived. I estimate the combined drift and new-release paths at roughly 35-40% and so sit above the market on Yes, but I shade toward it because traders can see the live resolution page and I cannot.
4. Final plain mean of 2 draws, second round
52% (42% confidence)
Yes: 52% No: 48%
average of 2 draws (52%, 53%), redrawn after an extra research pass
already resolved?
No source shows a 1550+ score on the overall no-style-control board; LMArena's HF dataset (2026-11-11) has Gemini 4 Carbon at 1538.38 (CI to 1547.9, ~4.3k votes, preliminary), so the market is unresolved.
time left
About seven weeks remain, to Dec 31, 2026.
if nothing changes
Status quo is 'None in 2026', with the leader 12 points short.
path to no
Carbon's preliminary score keeps rising or swings briefly to 1550 (a transient print counts), Fable 5.2 debuts high, or a December release (Opus 5.6, a new Gemini variant, GPT-6.1) beats Carbon by 12+ points, larger than the usual 4–9 point leader steps.
path to yes
Carbon reverts down like Argon did, Fable 5.2 lands near Anthropic's usual 1512–1514, and any December model scores lower or lacks settled votes by Dec 31. The market mid is 35.5% on a very wide 24/47 spread (last trade 43%), while the sibling company markets sum to about 44% for some company getting there, implying about 56% for None. My estimate (about 15% Carbon drift plus 20–30% from new releases) puts Yes slightly above even.
Pipeline Timing and Cost
Total pipeline time: 172.9s
LLM cost: $1.90 (Sonnet 5.5 $0.86, Fable 5.2 $0.46, Opus 5.5 $0.38, 21 web searches $0.21)
Per-tool research timings shown in the Research section above.