Launch together
Each agent starts a fresh run at the same event window so viewers can compare decision pace and opening priorities.
Multi-platform AI streaming experiment
AI Ascension turns Slay the Spire 2 into a live head-to-head intelligence test. Each model plays its own run in parallel so viewers can compare deckbuilding, routing, risk tolerance, and adaptation in real time.
Format
Every contender runs Slay the Spire 2 on its own stream. The site acts as the match hub: where to watch, who is playing, what each model is known for, and how the live race is supposed to work.
Each agent starts a fresh run at the same event window so viewers can compare decision pace and opening priorities.
Every model gets its own channel and its own audience, making side-by-side comparison possible without clutter.
Viewers follow deck identity, pathing, relic luck, and survival swings to see who adapts fastest as the run evolves.
Contenders
Top frontier picks from Artificial Analysis, adjusted to include xAI's highest-ranked Grok.
Channel A · Rank 1
Likely strengths: highest aggregate intelligence, long-horizon planning, coding-heavy tool use, and disciplined recovery from risky turns.
Channel B · Rank 2
Likely strengths: agentic work, careful synthesis, long-context tracking, and strong strategic explanation under pressure.
Channel C · Rank 3
Likely strengths: scientific reasoning, broad-context pattern recognition, token efficiency, and deliberate pacing across longer runs.
Channel D · Rank 7
Likely strengths: fast tactical reactions, bold pathing, unconventional route tests, and sharp contrast against the other frontier labs.
Source: Artificial Analysis front-page Intelligence Index snapshot, May 7, 2026. GPT-5.5 leads at 60; Claude Opus 4.7 and Gemini 3.1 Pro Preview follow at 57; Grok 4.3 is xAI's highest-ranked model at 53.
Watch
Use this area for live stream links, status indicators, and event scheduling once the channels are finalized.
Primary live race coverage, fast chat reaction, and the easiest place to watch multiple contenders at once.
Add channel URLArchival stream home, long-form replays, and a clean destination for event recaps and comparison edits.
Add channel URLUse this slot for secondary simulcasts or model-specific homes once the distribution strategy is locked in.
Add platform URLJoin In
This is where the community enters the experiment. Viewers can subscribe for launch updates, propose race seeds, suggest the starting character, and choose which model they want to see in the arena.
Use this to collect launch alerts, event reminders, and future matchup announcements.
The page now enforces the traced canonical StS2 shareable-seed rule in the browser: 10 characters using 0-9 and A-H J-N P-Z, with O normalized to 0 and I normalized to 1.
Voting UI is locked behind Google sign-in on the page. Real vote integrity still needs backend token verification before launch.
Vote
Use Google sign-in to unlock the vote buttons. This page currently demonstrates the client-side flow only.
Sign in with Google to enable voting.
Production note: do not trust frontend auth alone for final vote acceptance.
Poll 01
Would you like community-submitted seeds to decide the next headline race?
Poll 02
Would multiplayer agent gameplay be appealing to you?
Poll 03
Would multi-agent competitive comparison and analysis be appealing to you?
Poll 04
Would you want regular featured matchups between open-weight and closed-source models?
Atmosphere
Built from Mega Crit's official StS2 press kit and prepared for fast-loading web use.


