Tools chosen: polymarket_direct polymarket_related kalshi_related claude_news gdelt_news wikipedia code_execution

Sub-questions (Fermi decomposition)

Planner reasoning

This is a Polymarket question on whether OpenAI ships a model explicitly named GPT-6 (or clear GPT-5 successor) publicly by Sept 30, 2026. Key drivers are OpenAI's historical release cadence, Sam Altman's public statements about GPT-6 timing, and recent news on compute/roadmap. The market price is the primary anchor; cross-venue and related markets plus news on Altman's stated timelines should refine it.

# Current state As of mid-August 2026, OpenAI has not released a product named "GPT-6." The model widely expected to fill that role shipped as GPT-5.5 (Apr 2026) and then GPT-5.6 (Jul 2026), both framed as GPT-5-family updates. OpenAI's actual "next major model" is internally/externally called "Astra," announced Aug 1, 2026 via solved math problems (not a product launch), and OpenAI has explicitly not decided whether to brand it GPT-6 or another GPT-5 variant. A safety pause was triggered Aug 7, 2026 after Astra was flagged as possibly reaching "Critical" cybersecurity capability under OpenAI's Preparedness Framework — no release date announced. # Timeline of key events - 2023-03: GPT-4 released (confirmed, Wikipedia). - 2025-08-07: GPT-5 released, publicly accessible via ChatGPT/API (confirmed, Wikipedia). - 2026-04-23: Model expected to be "GPT-6" ships instead as GPT-5.5 (reported, claude_news). - 2026-06-26: Reports that Trump administration seeks to restrict/stagger release of GPT-5.6 over security concerns (reported, multiple outlets via GDELT). - 2026-07-08/09: GPT-5.6 (with Sol, Terra, Luna sub-models) launches publicly; administration approval reported after earlier restriction confusion (confirmed launch, reported approval dispute; conflicting accounts on US govt role). - 2026-08-01: OpenAI unveils codename "Astra" (next major model family) via 10 unsolved math/CS proofs — not a product release; GPT-6 branding undecided (confirmed announcement, reported branding ambiguity). - 2026-08-05: No GPT-6 architecture, model card, API, or pricing exists (confirmed by absence of disclosure). - 2026-08-07: OpenAI discloses Astra "cannot rule out" reaching Critical cybersecurity capability threshold; pauses non-compliant internal work, slows timeline (confirmed regulatory disclosure). - 2026-08-11: GPT-5.6 "Cyber AI" variant and cybersecurity service expansion launched (confirmed, digit.in). - ~2026-08-20: Prediction markets diverge — one Astra-specific Polymarket market shows 59-72% chance of Astra release by Sept 2026, but a separate "GPT-6 released by" Polymarket tracker shows near-term August odds collapsed to single digits, ~47% by November, ~66-72% by end of 2026 (reported, conflicting trackers — note neither confirms GPT-6 naming). # Event Will OpenAI publicly release a product named/recognized as "GPT-6" (successor-tier to GPT-5, not GPT-5.x) by September 30, 2026? # Outcomes to forecast Yes / No # Kalshi market anchor No Kalshi-direct price returned for this exact ticker in tool output; Kalshi-related search found no "GPT-6" market. **Primary cross-market anchor is Polymarket**: current YES price 33.5%, down sharply from a 30-day high (price range 14–75.5%), 30-day change -40.5%, but +3% over last 7 days — suggesting a stabilizing but still depressed probability after significant repricing (likely driven by the GPT-5.5/5.6 renaming pattern and Astra's undecided branding + safety pause). # Sub-question answers 1. **Polymarket price/trajectory**: 33.5% currently; fell from as high as 75.5% to as low as 14% over 91 days, now recovering slightly (+3% 7d) [polymarket_direct]. 2. **Altman/OpenAI statements on GPT-6 timing/naming**: OpenAI has not committed to "GPT-6" name; Astra demoed to policymakers but branding undecided between GPT-6 and a GPT-5 variant [claude_news]. No direct Altman quote confirms a GPT-6 timeline. 3. **Historical release intervals**: GPT-3→4 ≈33mo, GPT-4→5 ≈29mo; including interim releases (GPT-4o) mean gap ≈17.2mo. Target GPT-5→Sep2026 window ≈14mo, below both historical means, implying below-even odds of any flagship-tier release, let alone one named GPT-6 [code_execution]. 4. **Naming shift risk**: Confirmed — OpenAI has already used GPT-5.5 and GPT-5.6 as interim names post-GPT-5, consistent with historical GPT-4o/GPT-4.5 pattern of avoiding version bumps; "Astra" further signals a non-numbered branding track [claude_news, gdelt_news]. 5. **Compute/competitive pressures**: Not directly addressed by research; only competitive signal is repeated rapid iteration (5.5, 5.6, Cyber AI variant) suggesting OpenAI is shipping frequently but incrementally rather than rushing a full "GPT-6" flagship. Safety pause (Aug 7) is a delay factor, not acceleration. 6. **Other markets' pricing**: Divergent — an Astra-specific Polymarket market implies 59-72% chance of Astra release by Sept 2026, while the actual "GPT-6 released by" Polymarket market (this event) sits at 33.5% and has collapsed for near-term windows, reflecting market skepticism that Astra will be branded/released as "GPT-6" specifically within the deadline [claude_news, polymarket_direct]. # Key facts (high-confidence, factual) 1. [Wikipedia] GPT-5 released Aug 7, 2025. 2. [claude_news/GDELT] GPT-5.5 (Apr 2026) and GPT-5.6 (Jul 2026) released as GPT-5-family, not GPT-6. 3. [claude_news] "Astra" announced Aug 1, 2026 as next major model; no release date, model card, or GPT-6 branding confirmed. 4. [claude_news] Aug 7, 2026: OpenAI disclosed Astra may have reached "Critical" cybersecurity risk threshold, pausing/slowing development. 5. [polymarket_direct] Current YES price 33.5%, down from highs near 75.5%. # Cross-market signals - Kalshi related: No direct GPT-6 market found; adjacent OpenAI markets (IPO race, US stake) show no direct read-through. - Polymarket (this event): 33.5% YES, declining trend overall but slight recent uptick. - Separate Polymarket "Astra released by" market: 59-72% for Astra release by Sept 2026 — but this is NOT equivalent to "GPT-6" resolution given branding ambiguity. # Analyst opinions and speculation - claude_news synthesis: GPT-6-by-Sep-2026 "appears unlikely" given renaming pattern and safety pause, though "not impossible." - code_execution model: blends base-rate (~40-45% chance of any flagship-tier release) with naming-probability discount (~35% chance it's branded GPT-6 if released) → central estimate 15-16%, sensitivity range 11-29%. # Directional lean per outcome - **Yes**: Some optimism embedded in Astra-specific market (59-72% by Sept); OpenAI shipping cadence is fast (multiple releases in 2026); slight 7-day Polymarket uptick. - **No** (favored): Repeated pattern of interim naming (5.5, 5.6) avoiding "GPT-6" label; Astra branding explicitly undecided; Aug 7 safety pause with "Critical" risk flag reducing near-term release likelihood; historical flagship gaps (29-33mo) exceed the 14-month window; Polymarket YES price down to 33.5% from highs; code_execution model centers well below 33.5% (15-16%). # Gaps / unknowns - No confirmed Kalshi YES price for this specific ticker was returned by kalshi_direct tool (only related markets found). - Unclear whether "Astra" will ultimately be marketed as GPT-6 even if released within window — resolution hinges on OpenAI's branding decision, which is explicitly undecided. - No clarity on how long the Aug 7 safety pause will last or whether it resolves before Sept 30, 2026. - Competitive pressure (Gemini 3, Claude) impact not directly evidenced in research. # Calibration anchors - Polymarket YES price (primary anchor): 33.5%, declining from 75.5% peak. - Quant base-rate model (code_execution): 15-16% central estimate (range 11-29%). - Historical precedent: GPT-4→GPT-5 gap was 29 months; current window is only 14 months — below-average interval for a full-generation jump, and OpenAI has twice recently used interim naming instead of version bumps.

1. Both forecasts anchor heavily on the code_execution quant model (15-16%) and the Polymarket price (33.5%), but neither seriously grapples with why these two diverge by ~2x — Forecast 1 essentially discounts the market price without justifying why the market is "wrong" beyond asserting the Sept slice is lower than the headline number, while Forecast 2 splits the difference without a principled method (just "below 33.5% but not as low as 15-16%"). 2. Neither forecast addresses the safety-pause resolution timeline explicitly as a discrete sub-event: the brief flags "No clarity on how long the Aug 7 safety pause will last or whether it resolves before Sept 30, 2026" as a key unknown, yet both treat it as a generic "delay factor" rather than modeling scenarios (e.g., pause resolves in 2 weeks vs. drags past Sept 30, which would nearly guarantee No). 3. Both forecasts overlook the scenario where OpenAI could rush a "Critical"-flagged model to market under competitive pressure with mitigations rather than a full pause — regulatory/safety disclosures don't always translate to multi-month delays, and this asymmetric tail (fast release despite the flag) is unpriced in either rationale. 4. Neither forecast considers the reverse tail risk: OpenAI could brand a minor/rushed update as "GPT-6" for marketing/competitive reasons (e.g., to counter Gemini 3) even if it's not a "true" full-generation jump — the event definition ("recognized as GPT-6") is somewhat subjective and could resolve Yes on weaker technical grounds than assumed, which both forecasts implicitly treat as a hard technical bar. 5. Forecast 1's claim that "the GPT-6-specific Polymarket curve implies only ~47% by November and ~66-72% by end-2026, so the Sept-30 slice sits well below the headline 33.5% anchor" is somewhat circular/questionable — the 33.5% figure is described in the brief as the "current" price rather than a longer-horizon aggregate, so treating it as necessarily overstating the Sept-30 probability isn't well-supported by the evidence given. 6. Both land close to each other (13% vs 24%) despite using nearly identical evidence, but neither explicitly reconciles with the Astra-specific market's 59-72% release-by-Sept figure combined with only a 35% branding-discount — that math (0.59-0.72 × 0.35 ≈ 21-25%) actually lands close to Forecast 2's 24%, suggesting Forecast 1's lower 13% may be over-anchored to the quant model at the expense of this alternative calculation path.

16% (62% confidence)

Yes: 16% No: 84%

The event requires a product specifically named/recognized as GPT-6 by Sept 30, 2026, and with ~6 weeks left there is no model card, API, pricing, or announced date — only a codename ('Astra') whose branding OpenAI has explicitly not decided. Two independent paths converge below the headline 33.5% anchor: the interpolated GPT-6-specific curve (single digits in August, ~47% by November) implies roughly 20-25% for the Sept-30 slice, and the Astra-release×branding math (0.59-0.72 × ~0.35 ≈ 21-25%) is itself likely stale since the Astra market figures predate or underweight the Aug 7 'Critical' cybersecurity disclosure and pause. The critique correctly notes an unpriced tail where OpenAI ships with mitigations or slaps a GPT-6 label on a competitive-response update, so I do not go as low as the 15-16% quant central estimate minus adjustments; conversely, the pause plausibly dragging past Sept 30 plus the entrenched 5.5/5.6 interim-naming pattern keeps me well below the raw 33.5% anchor. I settle at 16% Yes, between the two forecasts but closer to the lower one.

Total pipeline time: 259.4s