← all forecasts
Will Google have the best AI model at the end of December 2026?
70%
Agent
72%
Market Price
-1.5pp
Edge
55%
Confidence
652,447
Volume
1.0c
Spread
49
Days to resolution
26
Markets in event
final rationale
- already resolved?
- Not resolved yet. Arena's official Hugging Face dataset is the 'text' config with no style control, published 9 Nov. It shows Google's Gemini 4 Carbon at #1 (1538) and Gemini 4 Argon at #2 (1530). Claude Opus 5.5 is third at 1513, about 25 points behind. The status quo therefore favours Google, but the check is a single snapshot on 31 Dec.
- time left
- About 7 weeks remain until the noon ET check on 31 Dec 2026.
- if nothing changes
- If nothing changes, Google wins: it holds two models above every rival, and the alphabetical tiebreak also favours Google.
- path to no
- Claude Fable 5.2, released 9 Nov and not yet on the Text table, posts a big jump. It already leads WebDev and the Agent boards, and a third-party Elo puts it at 1538. Alternatively, a further Anthropic or Meta release before year-end overtakes Carbon on the style-control-off board. Leadership changed hands 3–4 times in the past year.
- path to yes
- Fable 5.2 lands below Carbon, as Fable 5.1 sat about 23 points back, and no other frontier release beats Google. Google might also add another Gemini 4 variant, such as the rumoured Flash, a Pro or Deep Think. The market anchor is 71.5% and has been rising, which suggests traders have partly priced in the Fable 5.2 release. I shade slightly below the market because Fable 5.2's Text score is a large pending unknown and the window allows one more release cycle.
price vs. prediction
market price
agent forecast at t+0
10 follow-ups
earlier forecasts on this market
| date | predicted | market price | confidence |
|---|---|---|---|
| 2026-11-05 | 60% | 62% | 51% |
| 2026-10-29 | 55% | 52% | 50% |
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 resolution source is the key structured input, and the Polymarket price is the anchor. Related markets on both venues give cross-signals: sibling outcomes such as OpenAI, xAI and Anthropic, plus the Kalshi equivalents. One targeted search covers upcoming model releases (Gemini, GPT, Grok) expected before December 31.
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 Google have the best AI model at the end of December 2026?** - Current price (probability): 71.50% (live mid; bid 71% / ask 72%; last trade 72%) - 7-day price change: +8.00% - 30-day price change: +29.00% - Total volume: $652,447 (USD notional) - Price range: 6.00% - 71.50% - Data points: 87 days
- resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=2026-11-09 (leaderboard_publish_date); dataset repo last modified 2026-11-12. Arena's official leaderboard dataset (the 'text' config, which has no style control, 'latest' split, published 2026-11-09) ranks Google's Gemini 4 Carbon (High) #1 in the overall category with a score of 1538.28. Google's Gemini 4 Argon (High) is #2 at 1530.16
- polymarket_related: Polymarket search over 3 keyword(s), kept 18 matches. keyword 'best AI model end of 2026': 8 markets | keyword 'best AI model December': 8 markets | keyword 'top AI model': 2 markets
- kalshi_related: 2 related markets / summaries. keyword 'best AI model': ok | keyword 'LMArena': no matches | keyword 'top AI model': ok
- article_search: Fetched 16 unique articles across 2 queries (lookback=21d). 'LMArena leaderboard new top model November 2026': 8 articles | 'Gemini GPT Grok Claude release December 2026': 8 articles
Research loops, bottom lines
- current state: As of 9–12 Nov, Google holds #1 and #2 on Arena Text, with about a 15–20 point lead over the best rival (Meta's Muse Wonder, then Anthropic's Opus 5.5). That is a clear but not unassailable margin, since a new Anthropic, OpenAI or Meta release before 31 December could close it. I found no confirmed rival frontier release scheduled before the close.
- base rate: Google holds a roughly 15-point lead on the style-control-on mirror, and it holds #1 in the year's most recent snapshots. The year-end market is priced about 72% Google, up from about 40% a month or so ago. Leadership has swapped between Google and Anthropic several times in 2026, so an Anthropic release in the next seven weeks is the main risk; I could not check the official style-control-off table.
- yes case: The single most likely path to YES is Google simply staying put. Carbon and Argon already hold rank 1–2 by 15–20 points, and the Anthropic 9 Nov release shows no sign of closing that gap. No rival flagship is announced before the close, and a new Gemini 4 variant could only add to the lead.
- no case: The strongest NO case is Fable 5.2, released 9 Nov. Its Arena Text score is not yet visible, and Anthropic's agent and WebDev boards and third-party estimates already put it above Carbon. Google's roughly 15–20 point style-controlled lead has overlapping confidence intervals, and I could not verify the style-control-off ordering that actually resolves the market.
tools
| tool | status | time | items | summary |
|---|---|---|---|---|
| polymarket_direct | OK | 0.2s | 1 | ## This Market's Polymarket Data **Will Google have the best AI model at the end of December 2026?** - Current price (probability): 71.50% (live mid; bid 71% / ask 72%; last trade 72%) - 7-day price change: +8.00% - 30-day price change: +29.00% - Total volume: $652,447 (USD notional) - Price range: |
| resolution_source | OK | 24.6s | 1 | already_resolved=unclear; matches_criteria=True; last_updated=2026-11-09 (leaderboard_publish_date); dataset repo last modified 2026-11-12. Arena's official leaderboard dataset (the 'text' config, which has no style control, 'latest' split, published 2026-11-09) ranks Google's Gemini 4 Carbon (High) |
| polymarket_related | OK | 0.7s | 18 | Polymarket search over 3 keyword(s), kept 18 matches. keyword 'best AI model end of 2026': 8 markets | keyword 'best AI model December': 8 markets | keyword 'top AI model': 2 markets |
| kalshi_related | OK | 17.4s | 2 | 2 related markets / summaries. keyword 'best AI model': ok | keyword 'LMArena': no matches | keyword 'top AI model': ok |
| article_search | OK | 0.3s | 16 | Fetched 16 unique articles across 2 queries (lookback=21d). 'LMArena leaderboard new top model November 2026': 8 articles | 'Gemini GPT Grok Claude release December 2026': 8 articles |
web research agents
4/4 ok,
16 searches,
3 page reads
| dimension | status | time | searches | page reads | report |
|---|---|---|---|---|---|
| current state | OK | 22s | 3 | 1 | 3937 chars |
| base rate | OK | 30s | 3 | 2 | 5140 chars |
| yes case | OK | 18s | 4 | 0 | 4256 chars |
| no case | OK | 29s | 6 | 0 | 5777 chars |
report: current state
- [2026-11-12] BenchLeader mirrors the Arena dataset and shows Google's Gemini 4 Carbon (high) #1 on Arena Text at 1526. Gemini 4 Argon (high) is #2 at 1522, Meta's Muse Wonder #3 at 1511, and Claude Opus 5.5 #4 at 1506. Data is dated 2026-11-09. BenchLeader describes its figures as "with style control", but the market resolves on style control off. I could not confirm that setting's ranking. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-11-12] Google's lead over the best non-Google model is about 20 points (1526 against Muse Wonder 1511 and Opus 5.5 1506). Best Anthropic model is #4 (Opus 5.5, 1506). The best OpenAI model on that table is GPT-5.6 Sol at #22 (1485). The best xAI/SpaceXAI entry shown is Grok 4.20 at #38 (1475). Google, Meta and Anthropic hold the top 8 places. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-11-04] A second aggregator also shows Gemini 4 Carbon leading Text at 1,527.5. It notes the top confidence intervals overlap heavily, so the exact order is not reliable. https://17nas.com/llm-leaderboard.php
- [2026-10-28] Google announced Gemini 4 Carbon, calling it its strongest model. It ranks #1 on the Artificial Analysis Index at 59, ahead of Opus 5.5 (58). It had appeared in anonymous LMArena battles on 21 October. https://projedefteri.com/en/blog/gemini-4-carbon-release-date/
- [2026-09-30] Arena's X account said Gemini 4 Argon (High) landed #1 in Text Arena at 1525. That was about 20 points above #2 Claude Opus 4.6 (High). Google's earlier model was Gemini 3.8 Flash, at #11. https://x.com/arena/status/2105394855644139908
- [2026-11-09] On Arena's expert-prompt Text subcategory, Claude Opus 5.5 (High) is ranked 5 with a rank spread of 1–26 (1544±16). The wide spread shows how uncertain ranks are in the subcategories. https://arena.ai/leaderboard/chat/text/expert
- [2026-11-07] On the expert subcategory, a different aggregator shows Gemini 4 Argon at 1554 against Claude Opus 5 at 1553, a 1-point gap. https://llmbenchmark.ai/benchmarks/lmarena-text-expert
- [2026-11-11] Anthropic leads the Agent Arena, Code Agent and Work boards (Claude Fable 5.2 Max #1, Opus 5.5 #2). It also leads Arena WebDev (Opus 5.5, about 1810–1818). Neither board counts for this market, but both show Anthropic's strength. https://arena.ai/leaderboard/agent
- [2026-11-09] Arena's Search leaderboard is led by OpenAI's GPT-5.6 Sol (1256). It does not count for this market. https://www.benchleader.com/benchmarks/lmarena_search
- [2026-11-08] Unconfirmed leaks say Google is A/B testing a "Gemini 4 Flash" (codename Hydrogen) on LMArena. One X post claims a release around 18 November. Google has given no date. https://cellcog.ai/blog/gemini-4-flash-release-date/
- [2026-08] Claude Fable 5 was #1 at about 1525 in August, after Arena re-baselined its scores on 12 July. Google took #1 on 30 September with Gemini 4 Argon. Some aggregators, such as swfte.com, still show Claude on top. Their figures conflict with the Arena-sourced data and look stale or unreliable. https://localaimaster.com/blog/lmarena-chatbot-arena-leaderboard
- [2026-11-12] The Arena page itself (lmarena.ai, redirecting to arena.ai/leaderboard/text) blocked my access with a bot check and a 403. I could not read the official table directly. https://arena.ai/leaderboard/text
BOTTOM LINE: As of 9–12 Nov, Google holds #1 and #2 on Arena Text, with about a 15–20 point lead over the best rival (Meta's Muse Wonder, then Anthropic's Opus 5.5). That is a clear but not unassailable margin, since a new Anthropic, OpenAI or Meta release before 31 December could close it. I found no confirmed rival frontier release scheduled before the close.
NOT FOUND: A direct read of the official arena.ai table with style control off. Any confirmed scheduled rival model releases (OpenAI, Anthropic, xAI, Meta) for December 2026. The exact unrounded scores and rank spreads for the top models on the default view.
report: base rate
- [2026-11-12] Benchleader, which mirrors Arena's dataset and applies style control, had Google's Gemini 4 Carbon first at 1526 and Gemini 4 Argon second at 1522 (data dated 9 Nov). Meta's Muse Wonder is third at 1511 and Anthropic's best, Claude Opus 5.5, is fourth at 1506. OpenAI's best is GPT-5.6 Sol at 1485 (#22). Google's lead over the best non-Google model is about 15 points. This is a third-party mirror with style control ON, not the style-control-OFF table the market resolves on. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-11-12] I could not open arena.ai/leaderboard/text. It returned HTTP 403 and then a bot-check page. I could not verify the official table or the style-control-off ranking directly. https://arena.ai/leaderboard/text
- [2026-11-12] DeFi Rate, observed at 7:26 AM ET, showed Polymarket's "best AI model end of 2026" market at Google 72¢, Anthropic 22¢ and OpenAI 2.8¢. Google was #1 at the end of January and October on Polymarket's monthly markets. Anthropic was #1 at every month-end from February to September. The same page puts the end-of-November market at Google 79%. https://defirate.com/prediction-markets/best-ai-model-odds-this-year
- [2026-10-23 to 2026-10-01] The same market has swung hard. DeFi Rate's earlier snapshot, about 46 days before 12 Nov, had Anthropic 46% and Google 40%. On 1 Oct, Anthropic was 53% and Google 39.5%, with Google up 28 points in 24 hours. Prices flipped with the Gemini 4 launch, so the market reacts sharply to releases. https://polymarketer.com/odds/which-company-has-best-ai-model-end-of-2026/ (cited as polymarkettrader.com: https://polymarkettrader.com/odds/which-company-has-best-ai-model-end-of-2026/)
- [2026-11-11] A sibling Polymarket market, "best AI model on December 15", showed Google at about 69–75%. Volume was only $2.7K and some other outcomes looked mis-priced (StepFun 32%), so it is thin and noisy. https://polymarket.com/event/which-company-has-the-best-ai-model-on-december-15-2026
- [2026-11-07] llmbenchmark.ai's LMArena "expert" category had Gemini 4 Argon at 1554 and Claude Opus 5 at 1553, a one-point gap and effectively a tie. This is a sub-category, not the resolving table. https://llmbenchmark.ai/benchmarks/lmarena-text-expert
- [2026-11-12] Benchleader's weekly movement table shows that between 5 and 12 Nov, Anthropic models slipped (Opus 5.5 from #3 to #4, Fable 5.1 from #5 to #7). Meta's Muse Wonder rose from #4 to #3. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-11-04] An aggregator listed Gemini 4 Carbon as text #1 at 1527.5. It also noted that the top confidence intervals overlap heavily, so the exact order is not reliable. Carbon was also top on the Artificial Analysis Intelligence Index, while Claude Opus 5.5 led Arena's agent leaderboard. https://17nas.com/llm-leaderboard.php
- [2026-11 conflicting] swfte.com's "November 2026" page says Claude Mythos 5 is #1 at 1531, and it lists Gemini 3.1 Pro as a top model. It conflicts with the other trackers, which show Gemini 4 Carbon at #1 and say Mythos is only available through an invitation-only programme. Treat it as low reliability. https://www.swfte.com/lmarena
- [2025-12-31] Base rate for the year-end version of this question: Polymarket's "best AI model end of 2025" resolved Google (style control on, text arena), and its market ended at 100%. The 2025 market had $36M in volume. https://polymarket.com/event/which-company-has-best-ai-model-end-of-2025
- [2025-11-18] Google has repeatedly retaken #1 within days of a flagship launch. Gemini 3 Pro debuted at #1 with 1501, the first score above 1500, and Gemini 3.1 Pro (Feb 2026) tied for #1 at 1500, four points behind Opus 4.6. Anthropic held #1 at month-end from Feb to Sep 2026, so the lead has changed hands about 3–4 times in the last 12 months. https://x.com/arena/status/1990813759938703570 and https://x.com/arena/status/2024519891295089063?lang=en
- [2026-11-12] Leader changes tend to follow big releases. The main risks to Google before 31 Dec are an Anthropic release (the Opus/Fable line is near-tied with Google in some categories) or a Meta Muse update. A new release would need roughly 15 or more points of Arena score to retake rank 1, or to tie with overlapping confidence intervals. Under Arena's rank rules, models can share rank 1, and the tiebreak is then by score. https://www.benchleader.com/benchmarks/lmarena_text
BOTTOM LINE: Google holds a roughly 15-point lead on the style-control-on mirror, and it holds #1 in the year's most recent snapshots. The year-end market is priced about 72% Google, up from about 40% a month or so ago. Leadership has swapped between Google and Anthropic several times in 2026, so an Anthropic release in the next seven weeks is the main risk; I could not check the official style-control-off table.
NOT FOUND: the official arena.ai table (blocked); the style-control-OFF rank order; any announced Anthropic, OpenAI or Meta release dates before 31 Dec; any track record for the market's price as a forecaster; Kalshi's current Google price (it doesn't list Google, because Google has already been #1 this year).
report: yes case
YES means: Google's model holds rank 1 on the Arena Text leaderboard (style control off) on 31 Dec 2026, 12:00 PM ET. Ties on rank are broken by score, then alphabetically with Google ahead of other companies.
- [2026-11-12] I fetched BenchLeader's mirror of the Arena data, dated 9 Nov. Google holds ranks 1–2 (Gemini 4 Carbon 1526, Argon 1522). The next non-Google model is Meta's Muse Wonder at 1511, 15 points back. Claude Opus 5.5 is at 1506, and no Claude model is above 1506. Google's lead is about 15–20 points. This is a style-control-ON mirror. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-11-12] Google's lead over every rival is a margin on two separate models. Anthropic would have to pass both Carbon and Argon. The best OpenAI entry on the table is GPT-5.6 Sol at 1485 (#22). GPT-6.1 Sol is at 1482 (#27). OpenAI's September GPT-6 Astra release did not take the top spot. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-10-27] Carbon's 1528 on 27–28 Oct was effectively unchanged through 10 Nov (1527), so the lead is holding rather than decaying. Carbon had debuted at #1 on Text, #2 on Agent and #4 on WebDev. It is also #1 on the coding, hard-prompts, multi-turn and expert subcategories (aiblogly, read 10 Nov). Broad leads like this make a Text slip less likely. https://chatjudge.ai/benchmarks/chat/lmarena-text
- [2026-11-05] Carbon is #1 on Arena Coding at 1573, 22 points over Claude Opus 4.7/5.5 at 1551. Its lead is not limited to the overall table. https://www.benchleader.com/benchmarks/lmarena_text_coding
- [2026-11-09] The main rival release has already happened, with no visible effect on the table. Anthropic released Claude Fable 5.2 and Mythos 5.2 on 9 Nov (Fable 5.2 generally available). The table I fetched shows Fable 5.1 at 1503 and no Fable 5.2 entry. I did not find a published Arena score for Fable 5.2, so it is an unknown. Fable 5.1 trailed Carbon by 23 points. A 5.2 point release would need a large jump to pass. https://www.scriptbyai.com/anthropic-claude-timeline/
- [2026-11-12] Anthropic's cadence is point releases (5.5 family in Sep–Oct, 5.2 on 9 Nov). I found no Anthropic model announced beyond these. As of 31 Oct, none was named or dated (shattered.io). Haiku 5.5 shipped on 7 Oct (Anthropic newsroom). The window to 31 Dec is short, and Mythos 5.2 is restricted to approved organizations, so it would not necessarily be listed. https://www.anthropic.com/news
- [2026-11-12] Google's pipeline adds upside. A leaked "Gemini 4 Flash" test on Arena was reported on 8 Nov, but it is unconfirmed. Google went from announcing Argon (30 Sep) to Carbon (28 Oct) in four weeks. A third Gemini 4 variant, or a Pro or Deep Think version, could extend the lead. Gemini 3.8 Flash at 1498 (#9) shows even Google's mid-tier is near the top. https://en.wikipedia.org/wiki/Google_Gemini
- [2026-11-09] Meta's Muse Wonder (#3, 1511) is the only non-Anthropic challenger within range. Its gain from #4 to #3 in the past week is small, and it is still 15 points behind. https://www.benchleader.com/benchmarks/lmarena_text
- [2026-12-31 mechanism] Arena ranks use confidence intervals, so several top models can share rank 1. The rules then break ties by unrounded score, and the tiebreaker is Google first alphabetically. Google's 15-point score lead, with two models at the top, means it wins any such tie. A rival has to beat Google's score, not just overlap with it. https://www.benchleader.com/benchmarks/lmarena_text
- [2025-12-31] The pattern on year-end boards: Google's Gemini 3 Pro (Nov 2025) led at end-2025 and the end-of-2025 market resolved Google. Google has retaken #1 within days of each flagship launch. https://en.wikipedia.org/wiki/Google_Gemini
BOTTOM LINE: The single most likely path to YES is Google simply staying put. Carbon and Argon already hold rank 1–2 by 15–20 points, and the Anthropic 9 Nov release shows no sign of closing that gap. No rival flagship is announced before the close, and a new Gemini 4 variant could only add to the lead.
NOT FOUND: The official arena.ai table with style control off (blocked for colleagues, and I did not retry it). Fable 5.2's Arena score. Any announced OpenAI, Anthropic, Meta or xAI frontier release dates before 31 Dec.
report: no case
YES means: Google's model holds the top rank on the Arena Text leaderboard (style control off) at 12:00 PM ET on 31 Dec 2026, or wins the tiebreak on a shared rank 1.
- [2026-11-09] Claude Fable 5.2 and Mythos 5.2 were released on 9 Nov, three days before today. Fable 5.2 had already taken #1 on Arena WebDev on 10 Nov (1836±22). Arena's Text table snapshot of 9 Nov (Carbon 1526, Argon 1522) has no Fable 5.2 entry yet. Its Text score is therefore still pending. Anthropic's newest flagship is the largest unpriced threat to Google. https://arena.ai/leaderboard/code/webdev and https://www.scriptbyai.com/anthropic-claude-timeline/
- [2026-11-09] A third-party Elo site (openlm.ai, dated 9 Nov) puts Fable 5.2 first at 1538, ahead of Opus 5.5 at 1526 and Muse Wonder Full at 1518. Gemini 4 models do not appear in the lines I could read. This site says its Elo is computed by its own notebook, so it is not Arena's table and is weak evidence. It does show that third-party estimates already have Fable 5.2 above Google's 1526. https://openlm.ai/chatbot-arena/
- [2026-11-09] On the official Text page (Arena snapshot, search excerpt), Carbon is rank 1 at 1526±10 with rank spread 1–3, and Argon is rank 2 with spread 1–3. Carbon's own ±10 interval reaches down to about 1516. Muse Wonder (1509–1511) and Opus 5.5 (1506–1509) are about 15–20 points back, so the intervals do not clearly separate Google from them. Meta's rank spread starting at 1 is consistent with this. A 15-point gap can be closed by a single new model plus score drift. https://arena.ai/leaderboard/chat/text
- [2026-11-09] Fable 5.2 has not yet been shown on Arena's default Text table, so any claim of a Google lead through December depends on a score that does not exist yet. Anthropic held #1 at every month-end from February to September, and its Fable line was at about 1525 in August. New Anthropic models have launched about every 3–6 weeks (Opus 5.5 on 22 Sep, Haiku 5.5 on 7 Oct, Fable 5.2 on 9 Nov). A further Anthropic release before 31 Dec would not be unusual. Anthropic publishes no fixed schedule for its next model. https://github.com/jqueryscript/anthropic-claude-timeline and https://www.scriptbyai.com/anthropic-claude-timeline/
- [2026-11-09] On the Artificial Analysis Intelligence Index, Fable 5.2 scores 65 against 60 for Gemini 4 Carbon at launch (swfte.com, low-reliability aggregator). That reverses the 28 Oct ordering, when Carbon led at 59 against Opus 5.5 at 58. Intelligence-benchmark leadership has moved to Anthropic within two weeks of Carbon's launch. Arena Text is a preference vote, so it is only a loose predictor. https://www.swfte.com/ai/leaderboard
- [2026-11-11] The resolution wording is "style control off", but arena.ai shows style control as the default view. Arena made style control the default on 16 May 2025. A raw, non-style-controlled ranking exists separately in Arena's dataset ("text" config, not "text_style_control"). Every Google-leads figure in the reports comes from the style-controlled default or its mirrors. Without style control, rankings can shift between models with different verbosity and formatting. Whether Google still leads on the raw view is unverified. https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset
- [2026-10-26] OpenAI released GPT-6.1 Terra on 26 Oct and GPT-6.1 Sol earlier. OpenAI's best Arena Text models sit far from the top (#22 on the style-controlled mirror). Its Agent Arena placements are #6–#9. OpenAI's own text lead is unlikely by 31 Dec unless it ships a GPT-6.2-class model. OpenAI is not the main NO route. https://x.com/arena/status/2114783414771437852
- [2026-10-12] Meta released Muse Wonder on 12 Oct and Muse Wonder Full to Muse Code subscribers, with Ember in November (about monthly cadence). Wonder debuted at about 1509 (second place on 17 Oct). Reviewers have questioned how much its benchmark scores can be trusted, and Arena had a dispute with Meta over the Llama 4 submission. Meta's model could still be revised. Wonder Full is not in the API. It is unclear whether Arena's table lists it by default. https://www.mindstudio.ai/blog/meta-muse-wonder-launch and https://cryptobriefing.com/meta-muse-models-google-cloud/
- [2026-11-11] On Arena's Agent boards, Fable 5.2 (Max) leads, and Opus 5.5 is #2 at 19.5% against Fable's 30.35% on the overall board. Carbon was #2 on Agent Arena at launch on 28 Oct. Anthropic's Agent and Work boards show a large margin (Fable 5.2 beats Opus 5.5 by about 11 points on overall). That suggests Fable 5.2 is a step change over Opus 5.5. The Text arena may follow, but this is inference, not data. https://arena.ai/leaderboard/agent
- [2026-11-12] The question resolves at one instant (31 Dec, 12:00 PM ET), and leadership has flipped repeatedly. Anthropic held #1 on the Text table through the end of September, and Google regained it only on 30 Sep (Argon). Google's current lead is under six weeks old. Arena re-baselined scores on 12 July, so another methodology change before 31 Dec is possible. Ties on rank go to the higher score, so a Google rank-1 tie is not guaranteed. https://x.com/arena/status/2115495002998751262
BOTTOM LINE: The strongest NO case is Fable 5.2, released 9 Nov. Its Arena Text score is not yet visible, and Anthropic's agent and WebDev boards and third-party estimates already put it above Carbon. Google's roughly 15–20 point style-controlled lead has overlapping confidence intervals, and I could not verify the style-control-off ordering that actually resolves the market.
NOT FOUND: The official Arena Text table with style control off. Fable 5.2's Arena Text score. Any announced Anthropic, OpenAI, Meta or xAI frontier release for December. Evidence of a planned change to Arena's methodology before 31 Dec.
3. Forecasters 2 draws, Opus 5.5 + Fable 5.2
claude-opus-5-5, high effort
55% conf
| Yes | 69% |
| No | 31% |
- already resolved?
- Not resolved yet. Arena's official Hugging Face dataset is the 'text' config with no style control, published 9 Nov. It shows Google's Gemini 4 Carbon at #1 (1538) and Gemini 4 Argon at #2 (1530). Claude Opus 5.5 is third at 1513, about 25 points behind. The status quo therefore favours Google, but the check is a single snapshot on 31 Dec.
- time left
- About 7 weeks remain until the noon ET check on 31 Dec 2026.
- if nothing changes
- If nothing changes, Google wins: it holds two models above every rival, and the alphabetical tiebreak also favours Google.
- path to no
- Claude Fable 5.2, released 9 Nov and not yet on the Text table, posts a big jump. It already leads WebDev and the Agent boards, and a third-party Elo puts it at 1538. Alternatively, a further Anthropic or Meta release before year-end overtakes Carbon on the style-control-off board. Leadership changed hands 3–4 times in the past year.
- path to yes
- Fable 5.2 lands below Carbon, as Fable 5.1 sat about 23 points back, and no other frontier release beats Google. Google might also add another Gemini 4 variant, such as the rumoured Flash, a Pro or Deep Think. The market anchor is 71.5% and has been rising, which suggests traders have partly priced in the Fable 5.2 release. I shade slightly below the market because Fable 5.2's Text score is a large pending unknown and the window allows one more release cycle.
claude-fable-5-2, high effort
55% conf
| Yes | 71% |
| No | 29% |
- already resolved?
- Not resolved: the market checks the Arena Text table (style control off) at 12:00 PM ET on 31 Dec 2026, so Google must still hold first place then.
- time left
- About seven weeks remain.
- if nothing changes
- Status quo is YES: Arena's official Hugging Face dataset ('text' config, published 9 Nov) has Gemini 4 Carbon #1 at 1538.3, Gemini 4 Argon #2 at 1530.2 and Claude Opus 5.5 #3 at 1513.0. That is a roughly 25-point lead with two Google models on top, larger than the roughly 15 points on style-controlled mirrors, and the alphabetical tiebreak also favours Google.
- path to no
- Claude Fable 5.2, released 9 Nov, has no Text score yet. It already leads Arena WebDev and the Agent boards, and a low-reliability third-party Elo site puts it at 1538, so it could land above Carbon. A further Anthropic, Meta or OpenAI release, or a re-baselining or drift in Carbon's score on only 4,282 votes, could also flip a board that has changed hands three or four times in the past year.
- path to yes
- Fable 5.2 lands in the 1515-1535 range, short of the roughly 35-point jump over Fable 5.1 it would need. No rival flagship ships before year-end, or Google extends its lead with a Gemini 4 Flash or Pro variant. The market sits at 71.5% (up 29 points in 30 days, with Anthropic at 21%) and has had three days to price Fable 5.2. The larger style-control-off lead argues slightly up and the pending Fable 5.2 score slightly down, so I stay at the market price.
4. Final plain mean of 2 draws
70%
(55% confidence)
Yes: 70%
No: 30%
- already resolved?
- Not resolved yet. Arena's official Hugging Face dataset is the 'text' config with no style control, published 9 Nov. It shows Google's Gemini 4 Carbon at #1 (1538) and Gemini 4 Argon at #2 (1530). Claude Opus 5.5 is third at 1513, about 25 points behind. The status quo therefore favours Google, but the check is a single snapshot on 31 Dec.
- time left
- About 7 weeks remain until the noon ET check on 31 Dec 2026.
- if nothing changes
- If nothing changes, Google wins: it holds two models above every rival, and the alphabetical tiebreak also favours Google.
- path to no
- Claude Fable 5.2, released 9 Nov and not yet on the Text table, posts a big jump. It already leads WebDev and the Agent boards, and a third-party Elo puts it at 1538. Alternatively, a further Anthropic or Meta release before year-end overtakes Carbon on the style-control-off board. Leadership changed hands 3–4 times in the past year.
- path to yes
- Fable 5.2 lands below Carbon, as Fable 5.1 sat about 23 points back, and no other frontier release beats Google. Google might also add another Gemini 4 variant, such as the rumoured Flash, a Pro or Deep Think. The market anchor is 71.5% and has been rising, which suggests traders have partly priced in the Fable 5.2 release. I shade slightly below the market because Fable 5.2's Text score is a large pending unknown and the window allows one more release cycle.
Pipeline Timing and Cost
Total pipeline time: 73.3s
LLM cost: $1.37
(Sonnet 5.5 $0.68, Opus 5.5 $0.28, Fable 5.2 $0.23, 17 web searches $0.17)
Per-tool research timings shown in the Research section above.