Scan: Will Anthropic have the best Code Arena | WebDev AI at the end of September 2026?

date	predicted	market price	confidence
2026-09-03	79%	88%	52%
2026-08-27	85%	84%	74%
2026-08-20	77%	74%	64%

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct polymarket_related kalshi_related claude_news article_search gdelt_news wikipedia code_execution

Sub-questions (Fermi decomposition)

Which company's model currently holds rank 1 on the arena.ai Code Arena | WebDev (Overall, Models filter) leaderboard, and what is Anthropic's current rank and Elo gap to #1?
How volatile has the #1 position on the WebDev Arena been historically — how often has the leader changed in the past 6-12 months and how long has Anthropic held/lost the top spot?
What Anthropic model releases (e.g., next Claude Opus/Sonnet) are announced or credibly rumored to ship before September 30, 2026?
What competing releases from Google (Gemini), OpenAI (GPT), xAI (Grok), DeepSeek or others are expected before end of September 2026 that could displace Anthropic?
What do the sibling Polymarket markets (Google, OpenAI, xAI, Other) for this same event imply about the probability distribution after de-vigging?
Are there any changes to arena.ai's leaderboard methodology (style control, rank ties, model filtering) that could affect which model shows as rank 1?

Planner reasoning

This is a leaderboard-race question: the key facts are who currently tops the arena.ai Code Arena WebDev leaderboard, by what margin, and what model releases (Anthropic, Google, OpenAI, xAI, others) are expected before Sept 30, 2026. The Polymarket price is the primary anchor; sibling markets for other companies give a de-vigged distribution, and news on upcoming model launches drives the update.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.4s	1	## This Market's Polymarket Data Will Anthropic have the best Code Arena \| WebDev AI at the end of September 2026? - Current price (probability): 20.50% - 7-day price change: +6.00% - 30-day price change: -44.00% - Total volume: $53,647 (USD notional) - Price range: 12.50% - 89.00% - Data point
polymarket_related	OK	5.2s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'Code Arena WebDev': 0 markets \| keyword 'best AI model end of September': 0 markets \| keyword 'WebDev AI model': 0 markets
kalshi_related	OK	5.0s	0	0 related markets / summaries. keyword 'WebDev Arena': no matches \| keyword 'best AI model': no matches \| keyword 'Code Arena': no matches
claude_news	OK	21.7s	6	Based on my research, here are the key findings: - Current #1 as of early September 2026: OpenAI's GPT-6 Astra took the top spot on Code Arena's WebDev leaderboard with a score of 1,797, beating Anthropic's Claude Fable 5.1 (1,762) by 35 points, just two days after Astra's September 3, 2026 lau
article_search	OK	1.1s	31	Fetched 31 unique articles across 4 queries (lookback=45d). 'WebDev Arena leaderboard top model Claude': 10 articles \| 'arena.ai Code Arena WebDev ranking': 10 articles \| 'Anthropic Claude new model release 2026': 10 articles \| 'Gemini GPT Grok new model release coding benchmark 2026': 10 articles
gdelt_news	OK	205.0s	40	GDELT: 40 articles across 4 queries (lookback=30d). 'LMArena WebDev leaderboard': 10 hits \| 'Claude Opus release': 10 hits \| 'Gemini 3 release coding': 10 hits \| 'GPT-5 coding leaderboard': 10 hits
wikipedia	OK	0.2s	3	Fetched 3 Wikipedia entries (0 missing pages).
code_execution	OK	30.4s	0	Note on inputs: No live Polymarket quotes for this specific market group were passed into this session, so the figures below use illustrative-but-realistic placeholder prices reflecting current WebDev Arena sentiment (Anthropic currently favored). Swap in the live YES prices and re-run the de-vi

3. Evidence Brief Sonnet 5 · 6959 chars

# Current state As of the most recent confirmed leaderboard snapshots (Sept 3–8, 2026), OpenAI's GPT-6 Astra holds #1 on the Code Arena | WebDev leaderboard (~1,796–1,797), with Anthropic's best model (Claude Opus 5-max / Claude Fable 5.1) in #2, trailing by roughly 35–130 points depending on snapshot/version compared. Anthropic does NOT currently hold the top spot, though it held #1 briefly twice in the prior six weeks (late July with Claude Opus 5, then Sept 1–3 with Claude Fable 5.1) before being overtaken. # Timeline of key events - 2026-07-30 (confirmed): Claude Opus 5 leads WebDev Arena at ~1,704 Elo; Kimi K3 (open-weight) close behind (~29-pt gap). [ainexhub.com] - 2026-08-14 (confirmed): Google ships Gemini 3.7 Flash with coding/web-dev improvements; Google's flagship frontier model still not released. [fonearena.com, thestar.com.my] - 2026-09-01 (confirmed): Anthropic releases Claude Fable 5.1 and Mythos 5.1, described as strongest-ever cyber/coding capability; briefly takes #1 on WebDev leaderboard. [cbsnews.com, digg.com "Fable 5.1 Max Tops Code Arena WebDev," fonearena.com] - 2026-09-02 (confirmed): Google releases Gemini 3.8 Flash / Flash Cyber. [themobileindian.com, siliconangle.com] - 2026-09-03 (confirmed): OpenAI launches GPT-6 Astra; within two days it overtakes Claude Fable 5.1 for #1, scoring 1,797 vs. 1,762. [cryptobriefing.com, orcarouter.ai] - 2026-09-08 (confirmed): Pareto-view snapshot of arena.ai still shows OpenAI on top: gpt-6-astra-max 1,796 vs. claude-opus-5-max 1,688 (~108-pt gap). [arena.ai/leaderboard/code/webdev/pareto] # Event Will Anthropic own the #1-ranked model on the arena.ai Code Arena | WebDev (Overall, Models filter) leaderboard as checked Sept 30, 2026, 12:00 PM ET? # Outcomes to forecast Yes (Anthropic #1) / No (another company #1) # Kalshi market anchor No direct Kalshi price was returned (kalshi_related found 0 matching markets; no kalshi_direct quote provided). Primary cross-market anchor is therefore Polymarket: current YES (Anthropic) price = **20.5%**, up +6% over 7 days but down –44% over 30 days (range 12.5%–89%, $53.6k volume, 55 days of data). The 30-day plunge and 89% high-water mark align with Anthropic's brief #1 stint (Fable 5.1, Sept 1–3) before OpenAI's Astra overtook it. # Sub-question answers 1. **Current rank-1 holder / Anthropic's gap:** OpenAI (GPT-6 Astra, ~1,796–1,797) is #1 as of Sept 3–8, 2026; Anthropic (Claude Opus 5-max/Fable 5.1) is #2, trailing by ~35–130 Elo depending on snapshot. [claude_news synthesis, arena.ai pareto] 2. **Historical volatility of #1:** High — at least 3 changes in ~6 weeks (Opus 5 → Fable 5.1 → GPT-6 Astra), with the Fable 5.1 reign lasting only ~2 days before Astra overtook it. No official published "leadership duration" stats found; churn is qualitatively described as frequent. [tech-insider.org, cryptobriefing.com] 3. **Anthropic's pipeline:** No specific next-gen model (post-Fable 5.1/Mythos 5.1, e.g., "Opus 6") is confirmed or dated before Sept 30, 2026 in the research; Anthropic's release cadence (Opus 5→Fable 5.1 in ~5 weeks) suggests further releases plausible but nothing concrete is reported. 4. **Competing releases:** OpenAI's GPT-6 Astra (Sept 3) already displaced Anthropic. Google has shipped multiple Gemini 3.7/3.8 Flash variants (Aug 14–Sept 3) but its flagship frontier model remains undelivered per reporting ("Google shipped four Gemini Flash models... but flagship still nowhere to be seen," Sept 4). No Grok or DeepSeek release news found in this window. 5. **Sibling Polymarket markets / de-vig:** No real sibling market data was retrieved; the code_execution tool's de-vig (implying Anthropic ~57% fair probability) explicitly used illustrative placeholder prices, NOT live data — it should be disregarded as unreliable. The only real Polymarket price found is this market's own YES at 20.5%. 6. **Methodology changes:** No reported changes to arena.ai's WebDev leaderboard methodology (style control, tie rules) in the research window. # Key facts (high-confidence, factual) 1. [arena.ai pareto, Sept 8] gpt-6-astra-max (OpenAI) 1,796 vs claude-opus-5-max (Anthropic) 1,688. 2. [cryptobriefing.com] GPT-6 Astra (OpenAI) hit 1,797, topping Claude Fable 5.1 (1,762) just 2 days after its Sept 3 launch. 3. [fonearena.com, cbsnews.com] Anthropic released Claude Fable 5.1 and Mythos 5.1 on Sept 1, 2026, its most capable models yet. 4. [ainexhub.com] Late July 2026: Claude Opus 5 led at ~1,704, Kimi K3 close #2. 5. [polymarket_direct] This market's Anthropic-YES price: 20.5% now, was as high as 89% (likely during Fable 5.1's brief #1 run). # Cross-market signals - Kalshi related: none found (0 matches). - Polymarket (this market): 20.5% YES for Anthropic, sharp 44% 30-day decline reflecting loss of #1 to OpenAI. - Polymarket sibling markets (Google/OpenAI/xAI/Other): not retrieved; no real data available. - Sportsbook implied: n/a (not applicable to this event type). # Analyst opinions and speculation - claude_news synthesis: leadership is highly volatile (3 flips in ~6 weeks); Anthropic "could reclaim #1... plausible but not currently the case," citing its history of quick comebacks. - Illustrative Poisson model (code_execution, NOT based on real data): suggests only ~20% chance current leader (OpenAI) stays on top through Sept 30 if changeovers are memoryless — implying frequent flips could favor Anthropic reclaiming lead at some point, but this is explicitly flagged as illustrative/unreliable. # Directional lean per outcome - **Yes (Anthropic):** Supporting — demonstrated ability to retake #1 quickly (did so Sept 1); rapid release cadence; nearly 4 more weeks before resolution, ample time for a new Claude release. Opposing — currently trails by a real (~100+ pt) margin per most recent snapshot; OpenAI just launched a strong model with likely continued lead; Polymarket price for Anthropic fell sharply (44%) after losing #1. - **No (Other company, primarily OpenAI):** Supporting — OpenAI currently and recently confirmed #1 with a decisive Elo gap; Google also actively shipping (Gemini 3.8 Flash) though not yet frontier-topping. Opposing — history shows no single leader holds position long; Anthropic already proved capable of leapfrogging within the same window. # Gaps / unknowns - No live Kalshi YES price captured for this exact market. - No confirmed Anthropic roadmap/model announcement dated before Sept 30, 2026. - No real (non-illustrative) sibling Polymarket prices for Google/OpenAI/xAI/Other outcomes. - Unclear whether Astra's lead will hold, grow, or shrink through Sept 30 check date; no fresher (post-Sept 8) leaderboard data available. # Calibration anchors - Polymarket YES (Anthropic) — 20.5% (primary market-based anchor available). - Historical precedent: WebDev Arena #1 has changed at least 3 times in ~6 weeks (late July–Sept 2026), suggesting no persistent moat for any single lab over a ~4-week remaining horizon.

4. Forecasters Fable 5.1 + GPT-5.6 Sol · in parallel

claude-fable-5-1

52% conf

Yes	19%
No	81%

As of the latest snapshots (Sept 3–8), OpenAI's GPT-6 Astra leads Code Arena WebDev at ~1,796–1,797 while Anthropic's best trails by roughly 35–130 Elo, and Anthropic just spent its major release (Fable 5.1, Sept 1) with no confirmed follow-up model dated before Sept 30. Leadership has been volatile (3 flips in ~6 weeks), which keeps a comeback plausible, but a fresh Anthropic frontier release within ~3 weeks of its last one is unlikely and incremental score drift rarely closes a 100+ point gap. No Kalshi price exists, so I anchor on Polymarket's 20.5% YES and shade slightly lower given the sizable gap and Anthropic's recently exhausted release. Remaining uncertainty is mostly about an unannounced Claude launch or a leaderboard variant (e.g., a 'max' or thinking version) posting a surprise score.

gpt-5.6-sol

58% conf

Yes	21%
No	79%

The available market anchor is Polymarket’s 20.5% YES price; no direct Kalshi quote was available. OpenAI’s GPT-6 Astra currently leads Anthropic by roughly 35–130 Elo, pointing toward No because Anthropic must overcome a material gap by September 30. Frequent leadership changes and Anthropic’s two recent turns at #1 keep Yes plausible, but no post-Fable 5.1 Anthropic release is confirmed, while Astra is newly launched. I therefore stay close to the market consensus and assign only a slight upward adjustment for leaderboard volatility.

5. Devil's Advocate Sonnet 5 · red-team critique

Both forecasts anchor almost exactly on Polymarket's 20.5% without independently weighing the "historical precedent: #1 changed 3 times in 6 weeks" evidence, which if taken at face value (memoryless flip process) would imply a materially higher chance of at least one more flip favoring Anthropic before the Sept 30 checkpoint—neither forecaster explains why they discount this base-rate signal so heavily beyond calling the Poisson model "illustrative."

Neither forecast accounts for the possibility that OpenAI's lead could *grow* (Astra just launched and may see further tuning/"max" variants pushing the gap wider) versus shrink, which would bias toward even lower than 19-21% Yes; conversely neither considers Google's flagship frontier model (still undelivered per "Google shipped four Gemini Flash models... but flagship still nowhere to be seen") potentially entering and disrupting both OpenAI and Anthropic, which is irrelevant to Yes/No math here but shows incomplete competitive-landscape reasoning.

Both cite "no confirmed Anthropic roadmap release before Sept 30" as a key factor, but the brief explicitly notes Anthropic's cadence was Opus 5→Fable 5.1 in ~5 weeks, and Fable 5.1 launched Sept 1 — meaning a ~4-week resolution window is almost exactly consistent with another release landing before Sept 30; this cuts against their shared assumption that a new release is "unlikely," and neither forecaster flags this tension.

Neither forecaster addresses the reliability/staleness of the anchor data itself — the most recent leaderboard snapshot is Sept 8, over three weeks before resolution, yet both forecasts treat the 35-130 Elo gap as if it's current, without discounting for the ~22-day information gap during which multiple releases (per the observed ~2-week average flip rate) could easily occur.

Both forecasters converge to nearly identical probabilities (19% and 21%) despite theoretically independent reasoning, suggesting anchoring bias toward the Polymarket price rather than genuine independent synthesis of the volatility evidence; the brief's illustrative Poisson model (20% chance leader stays on top) and the qualitative "3 flips in 6 weeks" fact could support a wider range of outcomes (e.g., 25-35% Yes) that neither forecaster seriously explores.

Neither addresses potential resolution ambiguity — e.g., what happens if a tie or near-tie occurs, or if the "Overall, Models filter" methodology used for the Sept 30 check differs from the Pareto-view snapshots cited (Key fact 1 uses Pareto view, but the event resolution criterion doesn't specify Pareto vs Overall), which could introduce nontrivial rank differences given Anthropic's Opus 5-max score differs meaningfully between the two views (1,688 Pareto vs. other snapshots).

6. Aggregator Fable 5.1 · final calibrated forecast

22% (55% confidence)

Yes: 22% No: 78%

OpenAI's GPT-6 Astra holds #1 as of the latest (Sept 8) snapshot with a 35–130 Elo margin over Anthropic's best, and Anthropic just spent its major release (Fable 5.1) on Sept 1 with no confirmed follow-up before Sept 30. The observed leadership churn was release-driven rather than random drift, so the memoryless-flip argument overstates Anthropic's chances; Astra is freshly launched and could widen its lead. However, the critique is right that Anthropic's ~5-week cadence makes a late-September update or 'max'/thinking variant not implausible, and a ~35-point gap to the Fable 5.1 score is closable, so I nudge slightly above the two forecasts. With no Kalshi quote, Polymarket's 20.5% is the primary anchor, and 22% Yes reflects that anchor plus modest weight on Anthropic's demonstrated ability to retake #1 quickly.

Pipeline Timing

Total pipeline time: 299.0s

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