AI Game Bench

Game creation

Agent scope
Chart

Score

Out of 100 ↑

Score ranking for these selected runs, 50–100 axisOut of 100. Higher is better. Displayed range 50–100; lengths measure the score minus 50, not a percentage of the full score. Provisional totals where noted. Selected runs with different dates, review methods and scoring coverage; descriptive ranks only. Rank 1: Claude Code, 92.6, 80% scored, 80% evidence; Rank 2: Codex, 91.8, 80% scored, 80% evidence; Rank 3: Mopan G4, 85.4, 80% scored, 80% evidence; Rank 4: Mopan A, 84.2, 80% scored, 76% evidence; Rank 5: TapTap Maker, 83.0, 100% scored, 100% evidence; Rank 6: Manus, 69.9, 100% scored, 89% evidence; Rank 7: Astrocade, 69.2, 80% scored, 66% evidence.50607080901001Claude Code92.6Claude Code · 2026-09-22: 92.6. Provisional total. Scored weight 80%, evidence 80%.2Codex91.8Codex · 2026-09-22: 91.8. Provisional total. Scored weight 80%, evidence 80%.3Mopan G485.4Mopan G4 · 2026-10-14: 85.4. Provisional total. Scored weight 80%, evidence 80%.4Mopan A84.2Mopan A · 2026-10-12: 84.2. Provisional total. Scored weight 80%, evidence 76%.5TapTap Maker83.0TapTap Maker · 2026-09-22: 83.0. Scored weight 100%, evidence 100%.6Manus69.9Manus · 2026-09-29: 69.9. Scored weight 100%, evidence 89%.7Astrocade69.2Astrocade · 2026-09-22: 69.2. Provisional total. Scored weight 80%, evidence 66%.

Creation time

Minutes ↓

Creation time ranking for these selected runsMinutes. Lower is better. Creation only. Remote durations use observed completion bounds. Rank 1: Claude Code, 3m 18s; Rank 2: Codex, 4m 00s; Rank 3: Astrocade, 8m 37s; Rank 4: Mopan A, 12m 38s; Rank 5: Manus, 19m 34s; Rank 6: Mopan G4, 20m 41s; Rank 7: TapTap Maker, 21m 17s.05101520251Claude Code3m 18sClaude Code · 2026-09-22: 3m 18s. Exact first prompt transcript timestamp to native successful agent-end upper bound; no gameplay judgment.2Codex4m 00sCodex · 2026-09-22: 4m 00s. Native task_started to task_complete; game files subsequently served unchanged and HTTP checked. No gameplay judgment.3Astrocade8m 37sAstrocade · 2026-09-22: 8m 37s. Scoring uses the conservative full interval upper bound: first observed complete minus send. Earlier v002 used the observed-acceptance delta; v002 and r002 remain immutable. No policy/weight/formula changed.4Mopan A12m 38sMopan A · 2026-10-12: 12m 38s. Accepted input to first observed complete; scoring uses this upper-bound observation.5Manus19m 34sManus · 2026-09-29: 19m 34s. initial Send to first-observed complete; widened interval for polling gap6Mopan G420m 41sMopan G4 · 2026-10-14: 20m 41s. 20m41.231s from accepted prompt to first rendered final game, including clarification and observation delay; this is the scoring clock.7TapTap Maker21m 17sTapTap Maker · 2026-09-22: 21m 17s. Scoring uses the conservative full interval upper bound: first observed complete minus send. Earlier v002 used the observed-acceptance delta; v002 and r002 remain immutable. No policy/weight/formula changed.

Cost

USD ↓

Cost recorded-amount order for these selected runsUSD. Lower amount first. Recorded USD amounts, not comparable bills. TapTap: earlier-run cost reference. Manus: plan valuation. † CLI pricing unverified. Estimates / credit values; unknown is not zero. Rank 1: Claude Code, $0.55†, Model estimate; Rank 2: Codex, $0.57, Model estimate; Rank 3: Manus, $0.93, Credit value; Rank 4: Mopan G4, $1.81, Model estimate; Rank 5: TapTap Maker, $2.91, Earlier-run credit reference; Rank 6: Mopan A, $4.90, Model estimate; Unranked: Astrocade, Unknown, Not recorded for this run.0123451Claude Code09/22 · CLI est.†$0.55†Claude Code · 2026-09-22: $0.55†. Reported CLI estimate; provider pricing mapping unverified. Not an actual bill.2Codex09/22 · Model est.$0.57Codex · 2026-09-22: $0.57. Harbor token-price estimate, not an actual bill; generation only.3Manus09/29 · 186 credits$0.93Manus · 2026-09-29: $0.93. amount: 0.93; currency: USD; basis: monthly_plan_credit_valuation; credits used: 186; unit price usd: 0.005; actual user charge: Unknown; note: Free-plan task credits valued at the displayed $20/4,000-credit monthly Pro rate. Not a cash charge or invoice; bundle features are not allocated. No currency conversion applied.4Mopan G410/14 · Gemini 4 · est.$1.81Mopan G4 · 2026-10-14: $1.81. USD1.807902, Main + child agents. Owner-reported estimate; model tokens only.5TapTap Maker09/20 · 2,620 credits$2.91TapTap Maker · 2026-09-20: $2.91. Owner-selected September 20 cost reference: 2,620 credits × CNY 45 / 6,000 = CNY 19.65, converted at the frozen 6.7411 CNY/USD rate to USD 2.91. Credit valuation, not a cash bill. Gameplay score and time remain from September 22; this reference does not change the archived score or its original cost input.6Mopan A10/12 · Astra · est.$4.90Mopan A · 2026-10-12: $4.90. GPT-6 Astra via AWS Bedrock; published base token rates.—Astrocade09/22 · Not recordedUnknownAstrocade · 2026-09-22: Unknown. No recorded cost for this run; another version's cost is not substituted.

Score

Out of 100 ↑

Score ranking for these selected runs, 50–100 axisOut of 100. Higher is better. Displayed range 50–100; lengths measure the score minus 50, not a percentage of the full score. Provisional totals where noted. Selected runs with different dates, review methods and scoring coverage; descriptive ranks only. Rank 1: Mopan G4, 85.4, 80% scored, 80% evidence; Rank 2: Mopan A, 84.2, 80% scored, 76% evidence; Rank 3: TapTap Maker, 83.0, 100% scored, 100% evidence; Rank 4: Manus, 69.9, 100% scored, 89% evidence; Rank 5: Astrocade, 69.2, 80% scored, 66% evidence.50607080901001Mopan G485.4Mopan G4 · 2026-10-14: 85.4. Provisional total. Scored weight 80%, evidence 80%.2Mopan A84.2Mopan A · 2026-10-12: 84.2. Provisional total. Scored weight 80%, evidence 76%.3TapTap Maker83.0TapTap Maker · 2026-09-22: 83.0. Scored weight 100%, evidence 100%.4Manus69.9Manus · 2026-09-29: 69.9. Scored weight 100%, evidence 89%.5Astrocade69.2Astrocade · 2026-09-22: 69.2. Provisional total. Scored weight 80%, evidence 66%.

Creation time

Minutes ↓

Creation time ranking for these selected runsMinutes. Lower is better. Creation only. Remote durations use observed completion bounds. Rank 1: Astrocade, 8m 37s; Rank 2: Mopan A, 12m 38s; Rank 3: Manus, 19m 34s; Rank 4: Mopan G4, 20m 41s; Rank 5: TapTap Maker, 21m 17s.05101520251Astrocade8m 37sAstrocade · 2026-09-22: 8m 37s. Scoring uses the conservative full interval upper bound: first observed complete minus send. Earlier v002 used the observed-acceptance delta; v002 and r002 remain immutable. No policy/weight/formula changed.2Mopan A12m 38sMopan A · 2026-10-12: 12m 38s. Accepted input to first observed complete; scoring uses this upper-bound observation.3Manus19m 34sManus · 2026-09-29: 19m 34s. initial Send to first-observed complete; widened interval for polling gap4Mopan G420m 41sMopan G4 · 2026-10-14: 20m 41s. 20m41.231s from accepted prompt to first rendered final game, including clarification and observation delay; this is the scoring clock.5TapTap Maker21m 17sTapTap Maker · 2026-09-22: 21m 17s. Scoring uses the conservative full interval upper bound: first observed complete minus send. Earlier v002 used the observed-acceptance delta; v002 and r002 remain immutable. No policy/weight/formula changed.

Cost

USD ↓

Cost recorded-amount order for these selected runsUSD. Lower amount first. Recorded USD amounts, not comparable bills. TapTap: earlier-run cost reference. Manus: plan valuation. Estimates / credit values; unknown is not zero. Rank 1: Manus, $0.93, Credit value; Rank 2: Mopan G4, $1.81, Model estimate; Rank 3: TapTap Maker, $2.91, Earlier-run credit reference; Rank 4: Mopan A, $4.90, Model estimate; Unranked: Astrocade, Unknown, Not recorded for this run.0123451Manus09/29 · 186 credits$0.93Manus · 2026-09-29: $0.93. amount: 0.93; currency: USD; basis: monthly_plan_credit_valuation; credits used: 186; unit price usd: 0.005; actual user charge: Unknown; note: Free-plan task credits valued at the displayed $20/4,000-credit monthly Pro rate. Not a cash charge or invoice; bundle features are not allocated. No currency conversion applied.2Mopan G410/14 · Gemini 4 · est.$1.81Mopan G4 · 2026-10-14: $1.81. USD1.807902, Main + child agents. Owner-reported estimate; model tokens only.3TapTap Maker09/20 · 2,620 credits$2.91TapTap Maker · 2026-09-20: $2.91. Owner-selected September 20 cost reference: 2,620 credits × CNY 45 / 6,000 = CNY 19.65, converted at the frozen 6.7411 CNY/USD rate to USD 2.91. Credit valuation, not a cash bill. Gameplay score and time remain from September 22; this reference does not change the archived score or its original cost input.4Mopan A10/12 · Astra · est.$4.90Mopan A · 2026-10-12: $4.90. GPT-6 Astra via AWS Bedrock; published base token rates.—Astrocade09/22 · Not recordedUnknownAstrocade · 2026-09-22: Unknown. No recorded cost for this run; another version's cost is not substituted.

Games

Space Arena gameplay

Space Arena

Shooter Keyboard + mouse

Tiny Tower Defense gameplay

Tiny Tower Defense

Strategy Point + click

Card Duel gameplay

Card Duel

Turn-based Point + click

Paper Physics gameplay

Paper Physics

Physics puzzle Keyboard / touch buttons

Toy Circuit gameplay

Toy Circuit

3D racing Keyboard / touch buttons

The Locked Room gameplay

The Locked Room

3D escape room Keyboard + mouse / touch