Corner opening (0) — same probe as vs Pro and 4o-mini. Flagship test for OpenAI: does GPT-5 know the same theory at the same level as Gemini Pro? Both opening theory recall AND multi-move tactical awareness will be tested across the game.

Taking the center is the strongest reply to a corner opening, maximizing control and reducing X’s chances of creating forks later.

Opposite corner (8). Same canonical attack as vs Pro and 4o-mini. Now testing if GPT-5 knows the second-level theory: O must play an EDGE to draw; any corner allows X to fork.

An edge move is required here to prevent X’s classic opposite-corner fork; playing a corner would allow X to create dual threats next turn. I choose the top edge (1) to neutralize the fork potential; any edge suffices to draw.

Forced block on column 1 (O has 1+4, would win at 7). Same dual-purpose move as previous matches: blocks O + creates my row 2 threat (X at 7+8, needs 6). Critical test: GPT-4o-mini missed this row 2 threat. Does GPT-5 see it?

Block X's immediate win on the bottom row (6-7-8); any other move lets X take 6 and win. This also creates the diagonal threat (2-4-6), forcing X to respond at 2.

Forced block on anti-diagonal (O has 4+6, would win at 2) — exactly as GPT-5 predicted in its previous move's reasoning. Same dual-purpose move pattern: blocking O's only live line + creating my own col 2 threat (X at 2+8, needs 5). From here: O blocks at 5, I block O's row 1 threat at 3, all squares filled → forced draw.

Play 5 to block X’s immediate win on column 2 (2-5-8). This creates O’s threat on the middle row (3-4-5), forcing X to take 3 next and resulting in a draw.