Anthropic · model release forecast
When is the next Claude Haiku (4.6+) expected to be released?
Market-implied median · middle 50%: Oct 8 to Oct 25
44% within a week94% within a monthPrediction markets currently put the next Claude Haiku (4.6+) release around Oct 17, 2026, with a middle-50% window from Oct 8, 2026 to Oct 25, 2026. Markets imply a 44% chance of landing within a week of that date, and 94% within a month. The estimate is based on 4 active deadline contracts with $24K of traded volume. It has moved 9 days later over the past 7 days.
Release alerts
Get told when this date moves
One email the day before the the next Claude Haiku (4.6+) forecast date, and one on the day itself. Two emails per release, nothing else.
The evidence
Release probability by deadline
| Deadline | Probability bar | Probability |
|---|---|---|
| By Oct 7, 2026 | 23% | |
| By Oct 15, 2026 | 47% | |
| By Oct 31, 2026 | 92% | |
| By Dec 31, 2026 | 98% |
Forecast history
How this estimate has moved
Since Sep 29, 2026 the expected release date for the next Claude Haiku (4.6+) has moved 9 days later, from Oct 8, 2026 to Oct 17, 2026.
Common questions
About the Next Claude Haiku (4.6+) forecast
When is the next Claude Haiku (4.6+) expected to be released?
Prediction markets currently put the next Claude Haiku (4.6+) release around Oct 17, 2026, with a middle-50% window from Oct 8, 2026 to Oct 25, 2026. Markets imply a 44% chance of landing within a week of that date, and 94% within a month. The estimate is based on 4 active deadline contracts with $24K of traded volume. It has moved 9 days later over the past 7 days.
How precise is the Next Claude Haiku (4.6+) release date forecast?
Markets place 44% of the probability within a week of the median date and 94% within a month. A date with only a few percent inside a week should be read as a month, not a day.
Where does the Next Claude Haiku (4.6+) release forecast come from?
Each public "released by <date>?" contract on Polymarket gives one point on a cumulative probability curve. Release Oracle sorts those points, corrects small inconsistencies, then interpolates the 50% crossing for the headline date and the 25% and 75% crossings for its uncertainty window. Forecasts are market signals, not lab announcements, and can be wrong.
Also tracked