Model comparison

Results of the same prompt side by side. Nothing runs on this page – click a row to open the output. Sort by clicking a column header; the best value is highlighted.

⇆ Live side by side


Claude Sonnet 5.5 medium Anthropic Model profile →	loads	99	8/8	100	100 / 90	17 min 03 s	196,634	≈ $1.81 prices as of 06/10/2026	88 kB 1,648 lines
Claude Opus 5.5 medium Anthropic Model profile →	loads	97	8/8	99	100 / 80	67 min 54 s	214,420	≈ $3.95 prices as of 06/10/2026	106 kB 2,003 lines
Claude Haiku 5.5 medium Anthropic Model profile →	loads	96	8/8	100	100 / 70	31 min 13 s	234,719	≈ $0.11 prices as of 06/10/2026	54 kB 1,517 lines
GPT-6 Astra medium OpenAI Model profile →	loads	92	8/8	100	77 / 70	11 min 09 s	295,761	$2.06 prices as of 15/10/2026	38 kB 124 lines
GPT-6.1 Sol medium OpenAI Model profile →	loads	92	8/8	100	79 / 70	8 min 19 s	203,096	$0.26 prices as of 15/10/2026	49 kB 146 lines
GPT-5.5 medium OpenAI Model profile →	loads	90	7.5/8	98	100 / 60	1 min 51 s	11,362	$0.33 prices as of 17/10/2026	24 kB 321 lines
DeepSeek V4.1 Flash high DeepSeek Model profile →	loads	90	7.5/8	100	100 / 60	5 min 24 s	74,083	$0.044 prices as of 16/10/2026	79 kB 2,450 lines
Mistral Large 4 none Mistral Model profile →	loads	90	7.5/8	100	100 / 60	36 min 41 s	90,741	$0.19 prices as of 17/10/2026	65 kB 1,679 lines
Grok 4.6 default xAI Model profile →	loads	89	7.5/8	57	100 / 80	19 min 19 s	77,762	$0.46 prices as of 18/10/2026	66 kB 2,115 lines
Gemini 3.8 Flash medium Google Model profile →	loads	88	7/8	98	100 / 70	3 min 10 s	27,927	$0.10 prices as of 16/10/2026	68 kB 2,198 lines
GPT-5.6 Terra medium OpenAI Model profile →	loads	75	5.5/8	99	100 / 60	49 s	6,923	$0.079 prices as of 17/10/2026	17 kB 221 lines
GPT-5.6 Sol medium OpenAI Model profile →	loads	63	6/8	88	33 / 30	2 min 29 s	271,508	$0.34 prices as of 15/10/2026	21 kB 78 lines
Gemini 3.5 Flash-Lite medium Google Model profile →	loads	63	4/8	100	100 / 50	33 s	11,866	$0.029 prices as of 17/10/2026	31 kB 937 lines
GPT-6 Luna medium OpenAI Model profile →	loads	55	4.5/8	88	33 / 50	1 min 26 s	229,923	$0.00819 prices as of 15/10/2026	23 kB 56 lines
DeepSeek V4 Pro high DeepSeek Model profile →	loads	51	4/8	90	33 / 50	2 min 26 s	22,494	$0.044 prices as of 16/10/2026	54 kB 1,276 lines
Gemma 4 31B default Google Model profile →	loads	46	3.5/8	88	33 / 40	4 min 49 s	10,931	$0.00 prices as of 17/10/2026	31 kB 846 lines
Codestral 25.08 default Mistral Model profile →	loads	38	4/8	84	0 / 0	2 min 07 s	17,364	$0.015 prices as of 17/10/2026	81 kB 2,050 lines
Gemini 4 Argon medium Google Model profile →	loads	35	1.5/8	81	33 / 70	5 min 20 s	35,884	$0.35 prices as of 17/10/2026	98 kB 2,738 lines
Mistral Medium 3.5 high Mistral Model profile →	loads	26	2.5/8	76	0 / 0	4 min 09 s	40,966	$0.30 prices as of 17/10/2026	45 kB 1,510 lines
Gemini 3.1 Flash-Lite medium Google Model profile →	loads	12	0.5/8	83	0 / 0	8 s	2,957	$0.00385 prices as of 17/10/2026	5 kB 160 lines
Gemini 3.1 Pro medium Google Model profile →	loads	11	0.5/8	76	0 / 0	2 min 06 s	20,665	$0.24 prices as of 17/10/2026	51 kB 1,257 lines
GPT-6 Sol medium OpenAI Model profile →	broken fixed version →	0 after fix 86	0/8 after fix 7/8	70 after fix 99	0 / 0 after fix 89 / 70	3 min 11 s	441,896	$0.29 prices as of 15/10/2026	39 kB 215 lines
GPT-5.6 Luna medium OpenAI Model profile →	broken fixed version →	0 after fix 12	0/8 after fix 0/8	74 after fix 74	33 / 0 after fix 33 / 0	35 s	5,357	$0.00598 prices as of 17/10/2026	13 kB 101 lines
GPT-5.4 mini medium OpenAI Model profile →	broken fixed version →	0 after fix 7	0/8 after fix 0/8	69 after fix 69	0 / 0 after fix 0 / 0	7 min 09 s	108,599	$0.49 prices as of 17/10/2026	81 kB 2,547 lines
Mistral Small 4 high Mistral Model profile →	broken fixed version →	0 after fix 11	0/8 after fix 0.5/8	29 after fix 73	0 / 0 after fix 0 / 0	1 min 39 s	20,184	$0.012 prices as of 17/10/2026	30 kB 976 lines
GLM 5.3 high Z.ai Model profile →	broken fixed version →	0 after fix 67	0/8 after fix 4/8	69 after fix 99	0 / 0 after fix 100 / 80	7 min 11 s	73,508	$0.32 prices as of 17/10/2026	82 kB 1,884 lines
Gemini 3.6 Flash medium Google Model profile →	broken fixed version →	0 after fix 49	0/8 after fix 3.5/8	67 after fix 87	0 / 0 after fix 33 / 60	1 min 52 s	21,205	$0.078 prices as of 17/10/2026	42 kB 1,360 lines

Cost = tokens priced with the API price list valid on the date shown next to the cost (prices change). ≈ = estimate (only the total token count is known), no ≈ = exact from input / output tokens. “Loads” = the page opens without errors; technical health also checks behaviour during use – judge the quality in the viewer.

Evaluation

27 outputs from 27 models for the same zero-shot prompt – each one the model's first answer, published unedited.

4.2/8average score
21/27working outputs
85/100avg. technical health
50/100avg. UX

Broken (do not run): 6, 6 of them with a copy fixed by the same model. The average checklist score is 4.2/8 (a broken output scores 0), 5.4/8 among working ones. Best: Claude Opus 5.5, Claude Haiku 5.5, Claude Sonnet 5.5 and 2 more – 8/8. Weakest working: Gemini 3.1 Pro, Gemini 3.1 Flash-Lite – 0.5/8.

How we evaluate – methodology →

How the 8 requirements were met

From 21 working outputs: ✓ met, ½ partly, ✗ missing. Rate = (✓ + ½ × 0.5) / number of outputs.

Requirement	✓	½	✗	Rate
Procedural city Procedural city: roads, junctions, sidewalks, varied buildings, parks + another district; camera follows and zooms out with speed	9	10	2	67 %
Walk & run, enter and exit any car	12	3	6	64 %
3+ vehicle types with different handling, drift, collisions, visible damage	10	7	4	64 %
Traffic in lanes turning at junctions, pedestrians on sidewalks fleeing danger	8	8	5	57 %
Wanted 1–5 stars, escalating police, escaping lowers it, busted / wrecked costs money and respawns	10	7	4	64 %
2+ mission types with rewards, collectible pickups	11	7	3	69 %
HUD HUD: money, stars, health, objective with arrow, minimap	12	7	2	74 %
Web Audio sounds (engine, crash, siren), start screen with controls, pause menu	13	8	0	81 %

Strengths

The requirements the models met best – quotes from the reviewer's verdicts:

Web Audio sounds (engine, crash, siren), start screen with controls, pause menu · 81 % met

Synth engine, crash, siren sounds; start screen and pause menu. – Claude Opus 5.5
Synthesized engine/crash/siren, start screen and pause menu present. – Claude Haiku 5.5

HUD · 74 % met

HUD shows money, stars, health, objective arrow, minimap. – Claude Opus 5.5
HUD has money, stars, health, objective arrow and minimap. – Claude Haiku 5.5

Typical mistakes

The requirements the models most often missed or met only partly:

Traffic in lanes turning at junctions, pedestrians on sidewalks fleeing danger · 57 % met

Traffic is parked, pedestrians wander off sidewalks – Gemini 3.5 Flash-Lite
Traffic and pedestrian code has crash bugs – Mistral Medium 3.5

Wanted 1–5 stars, escalating police, escaping lowers it, busted / wrecked costs money and respawns · 64 % met

Police spawning triggers a crash – Mistral Medium 3.5
Wanted system coded, but unreachable in play. – Gemini 4 Argon

3+ vehicle types with different handling, drift, collisions, visible damage · 64 % met

Vehicle types exist, but driving throws an error – Mistral Medium 3.5
Cars, drift and damage coded but unreachable after crash. – Gemini 4 Argon

Agreement of two independent reviewers

The checklist of 27 outputs was filled in independently by two AI reviewers, Claude Opus 5.5 and Codex (GPT-6 Astra, medium). They agreed on 179 of 216 items (83 %). Average score by Claude 4.8/8, by Codex 4.5/8. Codex gave a lower score for 10, a higher one for 7 and the same for 10 outputs. They disagreed most often on: Walk & run, enter and exit any car (9×), Wanted 1–5 stars, escalating police, escaping lowers it, busted / wrecked costs money and respawns (7×), Web Audio sounds (engine, crash, siren), start screen with controls, pause menu (6×).

Where the reviewers disagree, the stricter (lower) verdict is proposed; the published score is approved by a human.

Objective tests

Measured by a tool, not by a person or an AI – so they are shown even before the scores are approved.

Tech: average 85/100 from 27 measurements; highest Claude Haiku 5.5 (100), lowest Mistral Small 4 (29). Measures Lighthouse, smoothness, errors during automated use, the mobile layout and whether the controls work.
UX / UI: average 50/100 from 27 measurements; highest Claude Opus 5.5 (100), lowest Gemini 3.1 Flash-Lite (0). Measures control size, layout at 3 widths, keyboard, feedback and readability.