Did GPT-6.1 Sol Actually Get Better? 3 Versions, 9 Games

GPT-6.1 Sol came out seven days after GPT-6 Sol. So I had three versions of OpenAI's Sol (GPT-5.6, GPT-6 and GPT-6.1) each build the same three games: Tetris, a 3D kart racer and a 3D shooter. Nine games, all made in the Codex app, from one request each, with nothing fixed afterwards. Then I played them. What I say about how they play is my personal opinion. Earlier videos: Tetris: https://youtu.be/e8MWhW_0ACw 3D kart racers: https://youtu.be/t61AbwOMcuQ OpenAI models in Claude Code, Claude models in Codex: https://youtu.be/JJWNl0PPQKU Chapters: 0:00 Intro 0:37 Tetris: GPT-5.6 Sol and GPT-6 Sol 1:12 Tetris: GPT-6.1 Sol's Lumen Stack 2:05 Kart: Turbo Drift Isle and Aster Circuit 2:35 Kart: GPT-6.1 Sol's Mesa Motors 3:09 The steering 3:36 Playing Mesa Motors 4:01 Shooter: Neon Bastion and Nightshift Outpost 4:34 Shooter: GPT-6.1 Sol's Riftwake 5:38 How long each build took 6:13 The price 6:38 Why so much like Astra? 7:40 So, did Sol get better? Play them yourself, free in your browser: Chromatic Tetris (GPT-5.6 Sol; "Chromatic Blocks" on the site): https://devarf.dev/play/chromatic-blocks/ Vector Stack (GPT-6 Sol): https://devarf.dev/play/vector-stack/ Lumen Stack (GPT-6.1 Sol): https://devarf.dev/play/lumen-stack/ Turbo Drift Isle (GPT-5.6 Sol): https://devarf.dev/play/turbo-drift-isle/ Aster Circuit (GPT-6 Sol): https://devarf.dev/play/aster-circuit/ Mesa Motors (GPT-6.1 Sol): https://devarf.dev/play/mesa-motors/ Neon Bastion (GPT-5.6 Sol): https://devarf.dev/play/neon-bastion-fps/ Nightshift Outpost (GPT-6 Sol): https://devarf.dev/play/nightshift-outpost-fps/ Riftwake (GPT-6.1 Sol): https://devarf.dev/play/riftwake-fps/ GPT-6 Astra's games, to compare: Ion Stack: https://devarf.dev/play/ion-stack/ Sunspur Kart: https://devarf.dev/play/sunspur-kart/ Amberline: https://devarf.dev/play/amberline-fps/ More games built with AI: https://devarf.dev How the games were made: Same user brief, default product settings, no custom skills, no added tools, no retries. All nine Sol games were built in the Codex desktop app, with no helpers. The prompt's first line was "Create a new folder, then build Tetris. Do your own research." (or "a 3D racing kart game", or "a 3D FPS shooter"), with the same list of rules under it. GPT-6 Sol and GPT-6.1 Sol got the same prompt in each game. I picked the same effort on every run. What each build ran on, with the effort label its log records: Chromatic Tetris: gpt-5.6-sol, Codex 0.147.0-alpha.6.5, Aug 2026, xhigh Turbo Drift Isle, Neon Bastion: gpt-5.6-sol, Codex 0.153.4, Sep 2026, max Vector Stack, Aster Circuit: gpt-6-sol, Codex 0.155.0-alpha.16, Sep 2026, max Nightshift Outpost: gpt-6-sol, Codex 0.159.0, Sep 2026, max Lumen Stack, Mesa Motors, Riftwake: gpt-6.1-sol, Codex 0.159.0, Sep 2026, max Astra's three: gpt-6-astra, Codex 0.153.4, Sep 2026, ultra (Ion Stack) and max The Codex app changed between most builds, not only the model, so what changed between versions isn't down to the model alone. Prices are what the tokens would cost at OpenAI's list price, read 30 September 2026.