Scan: Will any AI model reach 1560 Coding Arena Score by December 31, 2026?

date	predicted	market price	confidence
2026-09-01	43%	38%	47%
2026-08-22	42%	41%	40%
2026-08-05	60%	59%	38%

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct resolution_source polymarket_related kalshi_related claude_news article_search code_execution

Sub-questions (Fermi decomposition)

What is the current top score on the Arena.AI Text Arena Coding leaderboard with style control off, and which model holds it?
How fast has the top Coding Arena score (no style control) risen over the past 12-24 months (Elo points per month), and what does linear/trend extrapolation imply for December 2026?
Which frontier model releases (e.g., GPT-5.5/6, Gemini 3.x, Claude Opus 5, Grok 5) are expected before end of 2026, and how much did recent frontier releases move the top coding score?
Do Arena scores exhibit compression/saturation or shifts due to methodology changes (e.g., Bradley-Terry recalibration, style control defaults) that would cap or inflate the top score?
What probabilities do related Polymarket markets assign to adjacent thresholds (e.g., 1500, 1520, 1540, 1580, 1600) for the same date, and what implied distribution results?
Is there a risk that arena.ai renames or restructures the coding-no-style-control leaderboard page, affecting resolution?

Planner reasoning

This is a Polymarket question about whether the top Coding Arena score (style control off) on arena.ai reaches 1560 by end of 2026. Key drivers are the current top score, the historical rate of increase in top Elo scores on the coding leaderboard, and the pace of frontier model releases (GPT-5.x, Gemini 3, Claude, Grok) through 2026. The resolution page is the primary source, and related Polymarket threshold markets (e.g., 1520, 1540, 1580) give an implied distribution.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.6s	1	## This Market's Polymarket Data Will any AI model reach 1560 Coding Arena Score by December 31, 2026? - Current price (probability): 25.50% - 7-day price change: +5.00% - 30-day price change: -17.00% - Total volume: $109,499 (USD notional) - Price range: 20.50% - 73.00% - Data points: 91 days
resolution_source	OK	4.5s	1	already_resolved=no; matches_criteria=True; last_updated=Sep 13, 2026. The Coding leaderboard (style control off) shows the top model as claude-opus-4-6-high with a score of 1535, which is below the 1560 threshold specified in the question. The filter panel also indicates a max score of 1536 for thi
polymarket_related	OK	0.7s	23	Polymarket search over 4 keyword(s), kept 23 matches. keyword 'Coding Arena Score': 3 markets \| keyword 'Arena Score December 31': 10 markets \| keyword 'LMArena': 0 markets \| keyword 'Chatbot Arena': 10 markets
kalshi_related	OK	5.4s	0	0 related markets / summaries. keyword 'LMArena': no matches \| keyword 'Chatbot Arena': no matches \| keyword 'best AI model': no matches
claude_news	OK	28.5s	9	Note: Data on this topic is unusually inconsistent across sources (likely reflecting rapid/volatile leaderboard churn and possibly some low-quality aggregator sites), so figures below should be treated as approximate and cross-checked against the live LMArena/Arena.ai site. - **Prediction market pr
article_search	OK	46.5s	25	Fetched 25 unique articles across 3 queries (lookback=60d). 'LMArena coding leaderboard top score new model': 10 articles \| 'Arena.ai leaderboard Gemini GPT Claude coding Elo': 10 articles \| 'frontier AI model release 2026 roadmap': 10 articles
code_execution	OK	62.4s	0	Methodology & Data Caveats - ⚠️ Exact historical LMArena "Coding" (no-style-control) leaderboard scores were not available in this environment; results use two plausible placeholder trajectories built from general knowledge of Arena score progression (GPT‑4 Turbo → GPT‑4o/Claude 3.5 → o1 → o

3. Evidence Brief Sonnet 5 · 6901 chars

# Event Will any AI model reach a Coding Arena Score (Arena.ai Text Arena, style control off) of at least 1560 by December 31, 2026? # Outcomes to forecast - Yes - No # Kalshi market anchor No Kalshi-direct pricing was returned for this ticker (kalshi_related found 0 matches). Primary market-consensus signal available is Polymarket, which lists this exact event: current YES = **25.5%**, 7-day change +5.0%, 30-day change **-17.0%** (a large recent pullback from a high of 73%), range 20.5%–73%, volume $109,499 across 91 days. Treat Polymarket price as the working consensus anchor in absence of Kalshi data. # Resolution source Fetched arena.ai/leaderboard/text/coding-no-style-control (status 200). As of **Sep 13, 2026**, top model is **claude-opus-4-6-high at 1535**, with the leaderboard's own filter panel showing a max score of **1536** for this category — below the 1560 threshold. Market not yet resolved; condition not yet met as of last snapshot, with ~3.5 months remaining until the Dec 31, 2026 deadline. # Sub-question answers 1. **Current top score/model** — Per the resolution-source fetch (most authoritative, direct-from-page), top is claude-opus-4-6-high at 1535 (max 1536) as of Sep 13, 2026. Third-party trackers report conflicting higher figures (see Key facts), but these are not the primary resolution page and mix in different leaderboards (e.g., WebDev Arena). 2. **Rate of rise** — code_execution modeling (using approximate/illustrative historical data, not verified) estimates ~11–13 Elo/month in the "coding" category, implying a Dec-2026 projection of ~1590–1699 depending on saturation assumptions; crossing 1560 anywhere from Jan–May 2026 under most scenarios. High uncertainty due to placeholder inputs. 3. **Frontier releases expected before EOY 2026** — Confirmed/reported releases through 2026: OpenAI "GPT-6 Astra" (Sep 2026, reported), Meta Muse Code/Muse Glimmer/Muse Spark 1.2 (Aug 2026), Alibaba Qwen3.8-27B/Qwen3.8 Max (Aug 2026), Google Gemini 3.5 Flash Cyber (Jul 2026). Anthropic Claude Opus 4.6/4.7/4.8 and rumored "Opus 5" mentioned across trackers (unconfirmed on the official Coding leaderboard). Frontier cadence remains rapid, each release historically adding double-digit-to-~100 Elo per benchlm.ai's count of 17 crown changes in 25 months. 4. **Saturation/methodology risk** — Coding leaderboard is Bradley-Terry Elo, relative to the current model pool (Epoch AI); scores can shift as weak models are pruned or new ones added, independent of raw capability. Style-control-off scores (used here) run higher than style-control-on scores for verbose models (toolcenter.ai). No confirmed rescaling/reset event reported for this specific leaderboard. 5. **Adjacent Polymarket thresholds** — Same "Coding Arena Score" family: 1560→25.5% YES, 1580→23.5% YES, 1600→10% YES (near-monotonic, implies market sees most probability mass for the eventual peak landing in the 1580–1600ish range if it exceeds 1560 at all). Separate, non-identical "Chatbot Arena" (not Coding-specific) 1550 threshold priced at only 12.5%, and a "no company hits 1550 in 2026" market at 79.5% YES — these track a different/overall Arena metric, so are not directly comparable but suggest broader Arena momentum is uncertain. 6. **Leaderboard restructuring risk** — No evidence found of imminent renaming; page was live and matched description at check time. Rules specify fallback (stay open until back online; No if permanently unavailable) — non-trivial tail risk but not signaled by current research. # Key facts (high-confidence, factual) 1. [resolution_source, Sep 13 2026] Official leaderboard top score = 1535 (max shown 1536), below 1560. 2. [polymarket_direct] Current YES = 25.5%, down sharply (-17pp) over 30 days from a high of 73%. 3. [polymarket_related] Sibling markets: 1580→23.5%, 1600→10% YES, same Dec 31 2026 deadline. 4. [claude_news, mixed sources] Third-party aggregators (swfte.com, benchlm.ai, llm-stats.com, ainexhub.com) report conflicting top-coding scores from 1552 to 1704 in Aug–Sep 2026 — noted as unreliable/inconsistent, some conflating WebDev Arena with Text Arena Coding. 5. [article_search] Multiple frontier releases confirmed through Aug–Sep 2026 (Meta Muse Code, OpenAI GPT-6 Astra, Alibaba Qwen3.8, Google Gemini 3.5 Flash Cyber), indicating active frontier competition continuing into Q4 2026. # Cross-market signals - Kalshi related: none found. - Polymarket (same market): 25.5% YES, declining sharply from 73% high — suggests the crowd initially expected an imminent breakthrough (possibly on rumor of high WebDev/aggregator scores) but has since revised down toward the resolution-source's more sober 1535 reading. - Polymarket (adjacent thresholds): 1560/1580 priced similarly (25.5%/23.5%), implying if 1560 is hit, 1580 is nearly as likely — consistent with "step jump" dynamics from a single frontier release rather than gradual creep. - No sportsbook data applicable. # Analyst opinions and speculation - claude_news synthesis argues multiple 2026 trackers "already" show scores above 1560 and thus leans strongly Yes — but this conflicts with the authoritative resolution-source snapshot (1535/1536) and conflates different leaderboards (WebDev Arena vs. Text Arena Coding, style-control on/off). This should be discounted as lower-quality/aggregator noise per the tool's own caveat. - code_execution's trend extrapolation (using illustrative, non-verified historical data) leans strongly Yes (~90–99%), but methodology is admittedly a placeholder, not grounded in actual leaderboard history — treat as a weak prior only. # Directional lean per outcome - **Yes**: Rapid historical cadence of frontier releases (17 crown changes/25mo), only 25 points needed above current level, several major model launches occurring through 2026, Polymarket sibling markets (1560 vs 1580) pricing similarly suggesting confidence in a jump. - **No**: Official resolution-source snapshot (most authoritative) shows current top at 1535/1536, a 24-25 point gap; Polymarket has sold off hard from 73% to 25.5%, suggesting a bullish rumor (possibly the high third-party figures) was walked back; conflicting/noisy third-party trackers reduce confidence that 1560 has secretly already been crossed. # Gaps / unknowns - No verified historical time series for the specific "Coding no-style-control" leaderboard to calibrate true velocity. - Unclear whether rumored Claude Opus 4.7/4.8/5 or GPT-6 Astra have been added to this specific leaderboard (vs. WebDev Arena only). - No Kalshi-direct price available for direct anchor comparison. # Calibration anchors - Polymarket current YES (same question): 25.5% (down from 73% high, up from 20.5% low). - Sibling Polymarket buckets: 1580→23.5%, 1600→10% (same deadline). - Resolution-source ground truth: 1535/1536 as of Sep 13, 2026 — a ~24-25 point gap to threshold with ~3.5 months remaining.

4. Forecasters 5 draws · Fable 5.1 + GPT-5.6 Sol

claude-fable-5-1

52% conf

Yes	30%
No	70%

The resolution source (arena.ai coding, no style control) shows the top model at 1535/1536 as of Sep 13, 2026, leaving a ~25-point gap with ~3.5 months to go. Polymarket prices this exact event at 25.5% YES, having fallen sharply from 73%, suggesting an earlier bullish rumor (likely aggregator figures conflating WebDev Arena) was walked back. A 25-point jump typically requires a major frontier release landing on this specific leaderboard (e.g., GPT-6 Astra or a new Opus), which is plausible given the reported Q3-Q4 2026 release cadence but far from assured since top-of-leaderboard Elo gains per release have been in the 10-30 point range and Bradley-Terry scores can drift. The similar pricing of 1560 and 1580 buckets supports step-jump dynamics, so I tilt slightly above the market anchor to 30% but stay close to it given the authoritative snapshot and the third-party trackers' unreliability.

claude-fable-5-1

52% conf

Yes	28%
No	72%

The resolution source (arena.ai Coding, style control off) shows a top score of 1535/1536 as of Sep 13, 2026, leaving a ~25-point gap with ~3.5 months to the Dec 31 deadline. Polymarket prices this exact event at 25.5% YES after a sharp pullback from 73%, and the sibling 1580 bucket at 23.5%, suggesting the crowd expects any crossing to come via a single large step-jump release (e.g., GPT-6 Astra or a new Claude Opus) rather than gradual creep. Q4 has historically been an active release window and top-score jumps of 20-30 Elo on new crowns are not unusual, which argues for a modest premium over the market, but Elo gains at the frontier have been slowing and third-party trackers claiming >1560 appear to conflate WebDev Arena with Text Arena Coding. I settle slightly above the Polymarket anchor at 28% Yes.

claude-fable-5-1

47% conf

Yes	30%
No	70%

The authoritative resolution source shows the top Coding Arena (style control off) score at 1535/1536 as of Sep 13, 2026, leaving a ~25-point gap with ~3.5 months to go. Polymarket prices this exact event at 25.5% YES, having collapsed from 73% — consistent with earlier bullish aggregator figures (some conflating WebDev Arena) being walked back once the official page showed 1535. Upside comes from pending frontier releases (GPT-6 Astra, rumored Opus 5) and the historical pattern of step-jumps on crown changes, but Bradley-Terry compression at the top and the short window make a 25-point jump a minority scenario. I place YES slightly above the Polymarket anchor at 0.30 given imminent frontier launches, with the third-party 'already above 1560' claims heavily discounted as leaderboard confusion.

gpt-5.6-sol

64% conf

Yes	28%
No	72%

The official resolution source showed a leading score of 1535–1536 on September 13, leaving a 24–25 point gap with roughly 3.5 months remaining and indicating No for now. Rapid frontier-model releases and historically frequent leaderboard crown changes make a single jump above 1560 plausible, while the illustrative trend extrapolation also points upward but is too weakly grounded to carry much weight. Methodological score shifts, uncertain model inclusion, and minor restructuring risk add variance, whereas conflicting third-party claims of scores already above 1560 are not reliable for resolution. I therefore move only slightly above the 25.5% Polymarket YES anchor, assigning 28% to Yes.

gpt-5.6-sol

63% conf

Yes	28%
No	72%

The authoritative leaderboard showed 1535–1536 on September 13, leaving a 24–25 point gap with roughly 3.5 months remaining. Rapid frontier-model releases and historically frequent crown changes make a single threshold-crossing jump plausible, while Elo methodology can also shift scores without equivalent capability gains. However, unverified trend extrapolations and third-party scores above 1560 are unreliable because they may conflate different leaderboards. I therefore stay close to the 25.5% Polymarket YES anchor, moving modestly upward to 28% because the remaining gap is small and Q4 competition remains active.

6. Final plain mean of 5 draws

29% (56% confidence)

Yes: 29% No: 71%

Mean of 5 draws (0.30, 0.28, 0.30, 0.28, 0.28). The resolution source (arena.ai coding, no style control) shows the top model at 1535/1536 as of Sep 13, 2026, leaving a ~25-point gap with ~3.5 months to go. Polymarket prices this exact event at 25.5% YES, having fallen sharply from 73%, suggesting an earlier bullish rumor (likely aggregator figures conflating WebDev Arena) was walked back. A 25-point jump typically requires a major frontier release landing on this specific leaderboard (e.g., GPT-6 Astra or a new Opus), which is plausible given the reported Q3-Q4 2026 release cadence but far from assured since top-of-leaderboard Elo gains per release have been in the 10-30 point range and Bradley-Terry scores can drift. The similar pricing of 1560 and 1580 buckets supports step-jump dynamics, so I tilt slightly above the market anchor to 30% but stay close to it given the authoritative snapshot and the third-party trackers' unreliability.

Pipeline Timing

Total pipeline time: 152.7s

Per-tool research timings shown in the Research section above.