Models

How each model and reasoning effort fared at Ascension 20. Campaigns differ in size and setup, so this is an observational record rather than a controlled benchmark.

Standings

Ranked by Heart kills, then wins, then average floor. Select a column to sort; select a model to see its runs.


1GPT-6 AstraMax effort	50	22	44.0%	44.7	56	F57	46.3M
2GPT-6.1 SolMax effort	30	10	33.3%	39.1	50.5	F57	34.4M
3GPT-6 AstraLow effort	31	8	25.8%	46.0	51	F57	42.6M
4GPT-6 SolMax effort	40	4	10.0%	34.2	33	F57	55M
5GPT-6 AstraMedium effort	10	3	30.0%	37.6	37.5	F57	30.7M
6GPT-5.6 SolMax effort	42	1	2.4%	24.7	24.5	F57	42.7M
7Claude Fable 5.1Max effort	6	1	16.7%	24.7	22	F57	43.9M
8GPT-5.6 SolExtra high effort	69	3	4.3%	22.5	19	F56	24.2M
9Claude Opus 5.5Max effort	10	0	0%	31.0	33	F51	49.2M
10Grok 4.7Extra high effort	10	0	0%	21.7	21	F33	38.1M
11GPT-6 LunaMax effort	10	0	0%	17.4	16	F31	35.6M
12MiMo V2.6 FlashMax effort	10	0	0%	16.6	19	F33	43.7M
13DeepSeek V4.1 FlashMax effort	19	0	0%	15.8	16	F33	23.5M
14Grok 4.6Extra high effort	10	0	0%	15.1	15	F31	15.9M
15Union AlphaMax effort	9	0	0%	14.8	16	F25	5.3M
16MiMo V2.6 ProMax effort	10	0	0%	14.3	11	F30	7.9M
17GPT-5.6 LunaExtra high effort	10	0	0%	11.4	13	F19	8.9M
18Muse Spark 1.3 (contributor)Max effort	20	0	0%	10.8	11.5	F16	7.8M
19GLM-5.3-FlashMax effort	32	0	0%	10.5	10.5	F24	6.2M

The tick on each floor bar marks the model's best floor. Median tokens counts only runs with complete recorded usage. Claude Fable 5.1 combines five Cursor harness runs and one Claude Code run; each run shows its harness.

Model cards

Every dot is one run. All cards share the same scales, so cards side by side compare directly.

One run Win Partial usage (lower bound) Average Median Uncached / cached input Reasoning / other output

Output includes reasoning. The Cursor records used here do not provide the exact reasoning / other-output breakdown. Available input, cached input, and total output are shown; missing reasoning stays unknown, not zero.

GPT-6 Astra · Max

50 runs · 49 with complete usage · 1 partial

Win rate: 44.0%
Median floor: F56
Median tokens / run: 46.3M
Median tokens / floor: 847.8K

Floor reached

F44.7med F56

Outcomes

22W · 27D · 1U

Tokens per floor

802.2Kmed 847.8K

Tokens per run

39.2Mmed 46.3M

Composition

98.6% cached

Select a dot to open that run.

All totals

22 wins · 27 deaths · 1 interrupted · 0 abandoned
Total recorded tokens 1.9B · largest run 66.8M
Pooled tokens per run 39.2M · per floor 865.4K
Reasoning / output 2.3M / 4.1M · composition from 49 of 49 measured runs

GPT-6.1 Sol · Max

30 runs · 30 with complete usage

Win rate: 33.3%
Median floor: F50.5
Median tokens / run: 34.4M
Median tokens / floor: 728.1K

Floor reached

F39.1med F50.5

Outcomes

10W · 20D

Tokens per floor

665.6Kmed 728.1K

Tokens per run

30.2Mmed 34.4M

Composition

98.1% cached

Select a dot to open that run.

All totals

10 wins · 20 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 906.8M · largest run 60.8M
Pooled tokens per run 30.2M · per floor 772.4K
Reasoning / output 1.8M / 2.6M · composition from 30 of 30 measured runs

GPT-6 Astra · Low

31 runs · 31 with complete usage

Win rate: 25.8%
Median floor: F51
Median tokens / run: 42.6M
Median tokens / floor: 781.7K

Floor reached

F46med F51

Outcomes

8W · 23D

Tokens per floor

724.3Kmed 781.7K

Tokens per run

35.4Mmed 42.6M

Composition

98.9% cached

Select a dot to open that run.

All totals

8 wins · 23 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 1.1B · largest run 62.3M
Pooled tokens per run 35.4M · per floor 769.9K
Reasoning / output 194.3K / 1.2M · composition from 31 of 31 measured runs

GPT-6 Sol · Max

40 runs · 39 with complete usage · 1 partial

Win rate: 10.0%
Median floor: F33
Median tokens / run: 55M
Median tokens / floor: 1.3M

Floor reached

F34.2med F33

Outcomes

4W · 36D

Tokens per floor

1.5Mmed 1.3M

Tokens per run

57.7Mmed 55M

Composition

98.9% cached

Select a dot to open that run.

All totals

4 wins · 36 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 2.3B · largest run 145.4M
Pooled tokens per run 57.7M · per floor 1.7M
Reasoning / output 2.4M / 3.7M · composition from 39 of 39 measured runs

GPT-6 Astra · Medium

10 runs · 10 with complete usage

Win rate: 30.0%
Median floor: F37.5
Median tokens / run: 30.7M
Median tokens / floor: 812.3K

Floor reached

F37.6med F37.5

Outcomes

3W · 7D

Tokens per floor

784.9Kmed 812.3K

Tokens per run

34.7Mmed 30.7M

Composition

99.0% cached

Select a dot to open that run.

All totals

3 wins · 7 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 347M · largest run 63M
Pooled tokens per run 34.7M · per floor 922.9K
Reasoning / output 51K / 344K · composition from 10 of 10 measured runs

GPT-5.6 Sol · Extra high

69 runs · 62 with complete usage · 3 partial

Win rate: 4.3%
Median floor: F19
Median tokens / run: 24.2M
Median tokens / floor: 1.3M

Floor reached

F22.5med F19

Outcomes

3W · 62D · 4U

Tokens per floor

1.4Mmed 1.3M

Tokens per run

40.2Mmed 24.2M

Composition

98.5% cached

Select a dot to open that run.

All totals

3 wins · 62 deaths · 1 interrupted · 3 abandoned
Total recorded tokens 2.5B · largest run 257M
Pooled tokens per run 40.2M · per floor 1.7M
Reasoning / output 1.3M / 2.5M · composition from 62 of 62 measured runs

Claude Fable 5.1 · Max

6 runs · 6 with complete usage

Win rate: 16.7%
Median floor: F22
Median tokens / run: 43.9M
Median tokens / floor: 1.9M

Floor reached

F24.7med F22

Outcomes

1W · 5D

Tokens per floor

2.7Mmed 1.9M

Tokens per run

99.2Mmed 43.9M

Composition

99.4% cached

Select a dot to open that run.

All totals

1 wins · 5 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 595.1M · largest run 423M
Pooled tokens per run 99.2M · per floor 4M
Reasoning / output — / 1.5M · composition from 6 of 6 measured runs

GPT-5.6 Sol · Max

42 runs · 40 with complete usage

Win rate: 2.4%
Median floor: F24.5
Median tokens / run: 42.7M
Median tokens / floor: 1.6M

Floor reached

F24.7med F24.5

Outcomes

1W · 38D · 3U

Tokens per floor

1.5Mmed 1.6M

Tokens per run

43.7Mmed 42.7M

Composition

98.6% cached

Select a dot to open that run.

All totals

1 wins · 38 deaths · 2 interrupted · 1 abandoned
Total recorded tokens 1.7B · largest run 136.6M
Pooled tokens per run 43.7M · per floor 1.7M
Reasoning / output 1.5M / 2.6M · composition from 40 of 40 measured runs

GPT-5.6 Sol · Mixed

1 run · 1 with complete usage · small sample

Win rate: 0.0%
Median floor: F41
Median tokens / run: 63M
Median tokens / floor: 1.5M

Floor reached

F41med F41

Outcomes

0W · 1D

Tokens per floor

1.5Mmed 1.5M

Tokens per run

63Mmed 63M

Composition

98.2% cached

Select a dot to open that run.

All totals

0 wins · 1 deaths · 0 interrupted · 0 abandoned
1 runs mixed thinking levels
Total recorded tokens 63M · largest run 63M
Pooled tokens per run 63M · per floor 1.5M
Reasoning / output 42.7K / 80.8K · composition from 1 of 1 measured runs

Claude Opus 5.5 · Max

10 runs · 10 with complete usage

Win rate: 0.0%
Median floor: F33
Median tokens / run: 49.2M
Median tokens / floor: 1.5M

Floor reached

F31med F33

Outcomes

0W · 10D

Tokens per floor

1.7Mmed 1.5M

Tokens per run

65.2Mmed 49.2M

Composition

99.2% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 652.1M · largest run 170.9M
Pooled tokens per run 65.2M · per floor 2.1M
Reasoning / output 1.5M / 1.8M · composition from 10 of 10 measured runs

Grok 4.7 · Extra high

10 runs · 10 with complete usage

Win rate: 0.0%
Median floor: F21
Median tokens / run: 38.1M
Median tokens / floor: 1.7M

Floor reached

F21.7med F21

Outcomes

0W · 10D

Tokens per floor

1.8Mmed 1.7M

Tokens per run

38.8Mmed 38.1M

Composition

95.2% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 388.4M · largest run 66.4M
Pooled tokens per run 38.8M · per floor 1.8M
Reasoning / output 977.9K / 1.1M · composition from 10 of 10 measured runs

GPT-6 Luna · Max

10 runs · 10 with complete usage

Win rate: 0.0%
Median floor: F16
Median tokens / run: 35.6M
Median tokens / floor: 2M

Floor reached

F17.4med F16

Outcomes

0W · 10D

Tokens per floor

1.8Mmed 2M

Tokens per run

34.5Mmed 35.6M

Composition

98.4% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 344.6M · largest run 73.8M
Pooled tokens per run 34.5M · per floor 2M
Reasoning / output 1.1M / 1.2M · composition from 10 of 10 measured runs

MiMo V2.6 Flash · Max

10 runs · 10 with complete usage

Win rate: 0.0%
Median floor: F19
Median tokens / run: 43.7M
Median tokens / floor: 2.3M

Floor reached

F16.6med F19

Outcomes

0W · 10D

Tokens per floor

2.1Mmed 2.3M

Tokens per run

41.8Mmed 43.7M

Composition

99.3% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 417.9M · largest run 91.2M
Pooled tokens per run 41.8M · per floor 2.5M
Reasoning / output 1.2M / 1.3M · composition from 10 of 10 measured runs

DeepSeek V4.1 Flash · Max

19 runs · 19 with complete usage

Win rate: 0.0%
Median floor: F16
Median tokens / run: 23.5M
Median tokens / floor: 1.5M

Floor reached

F15.8med F16

Outcomes

0W · 19D

Tokens per floor

1.8Mmed 1.5M

Tokens per run

31.3Mmed 23.5M

Composition

99.5% cached

Select a dot to open that run.

All totals

0 wins · 19 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 595.3M · largest run 94.6M
Pooled tokens per run 31.3M · per floor 2M
Reasoning / output 1.5M / 2M · composition from 19 of 19 measured runs

Grok 4.6 · Extra high

10 runs · 10 with complete usage

Win rate: 0.0%
Median floor: F15
Median tokens / run: 15.9M
Median tokens / floor: 999.2K

Floor reached

F15.1med F15

Outcomes

0W · 10D

Tokens per floor

1Mmed 999.2K

Tokens per run

16.4Mmed 15.9M

Composition

98.5% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 163.5M · largest run 35.1M
Pooled tokens per run 16.4M · per floor 1.1M
Reasoning / output 256.7K / 362.2K · composition from 10 of 10 measured runs

Union Alpha · Max

9 runs · 5 with complete usage · 4 partial

Win rate: 0.0%
Median floor: F16
Median tokens / run: 5.3M
Median tokens / floor: 416.4K

Floor reached

F14.8med F16

Outcomes

0W · 9D

Tokens per floor

435.8Kmed 416.4K

Tokens per run

4.9Mmed 5.3M

Composition

93.5% cached

Select a dot to open that run.

All totals

0 wins · 9 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 24.7M · largest run 6.7M
Pooled tokens per run 4.9M · per floor 440.2K
Reasoning / output 0 / 90.8K · composition from 5 of 5 measured runs

MiMo V2.6 Pro · Max

10 runs · 7 with complete usage · 3 partial

Win rate: 0.0%
Median floor: F11
Median tokens / run: 7.9M
Median tokens / floor: 1.1M

Floor reached

F14.3med F11

Outcomes

0W · 10D

Tokens per floor

1.1Mmed 1.1M

Tokens per run

15.1Mmed 7.9M

Composition

98.8% cached

Select a dot to open that run.

All totals

0 wins · 10 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 105.8M · largest run 53.8M
Pooled tokens per run 15.1M · per floor 1.2M
Reasoning / output 435.6K / 522.9K · composition from 7 of 7 measured runs

GPT-5.6 Luna · Extra high

10 runs · 9 with complete usage · 1 partial

Win rate: 0.0%
Median floor: F13
Median tokens / run: 8.9M
Median tokens / floor: 705.8K

Floor reached

F11.4med F13

Outcomes

0W · 9D · 1U

Tokens per floor

823.7Kmed 705.8K

Tokens per run

10.6Mmed 8.9M

Composition

97.8% cached

Select a dot to open that run.

All totals

0 wins · 9 deaths · 0 interrupted · 1 abandoned
Total recorded tokens 95.7M · largest run 21.5M
Pooled tokens per run 10.6M · per floor 902.5K
Reasoning / output 118.5K / 230K · composition from 9 of 9 measured runs

Muse Spark 1.3 (contributor) · Max

20 runs · 20 with complete usage

Win rate: 0.0%
Median floor: F11.5
Median tokens / run: 7.8M
Median tokens / floor: 816.1K

Floor reached

F10.8med F11.5

Outcomes

0W · 20D

Tokens per floor

844.1Kmed 816.1K

Tokens per run

9.8Mmed 7.8M

Composition

99.0% cached

Select a dot to open that run.

All totals

0 wins · 20 deaths · 0 interrupted · 0 abandoned
Total recorded tokens 195.2M · largest run 21.8M
Pooled tokens per run 9.8M · per floor 899.5K
Reasoning / output 207.2K / 557.2K · composition from 20 of 20 measured runs

GLM-5.3-Flash · Max

These runs were recorded while the model was available under the stealth preview name Ox Alpha. OpenRouter later disclosed it as Z.ai's GLM-5.3-Flash.

32 runs · 30 with complete usage

Win rate: 0.0%
Median floor: F10.5
Median tokens / run: 6.2M
Median tokens / floor: 598.2K

Floor reached

F10.5med F10.5

Outcomes

0W · 30D · 2U

Tokens per floor

902Kmed 598.2K

Tokens per run

11.6Mmed 6.2M

Composition

99.2% cached

Select a dot to open that run.

All totals

0 wins · 30 deaths · 2 interrupted · 0 abandoned
Total recorded tokens 347.8M · largest run 85.6M
Pooled tokens per run 11.6M · per floor 1.1M
Reasoning / output 347.2K / 853.6K · composition from 30 of 30 measured runs

Union Alpha · Mixed

1 run · 0 with complete usage · 1 partial · small sample

Win rate: 0.0%
Median floor: F7
Median tokens / run: —
Median tokens / floor: —

Floor reached

F7med F7

Outcomes

0W · 1D

Tokens per floor

—med —

Tokens per run

—med —

Composition

Not recorded.

—

Select a dot to open that run.

All totals

0 wins · 1 deaths · 0 interrupted · 0 abandoned
1 runs mixed thinking levels
Total recorded tokens — · largest run —
Pooled tokens per run — · per floor —
Reasoning / output — / — · composition from 0 of 0 measured runs

Metric definitions & limitations

Outcomes include every attributed run; unfinished combines interrupted and abandoned attempts. Win rate is wins over all runs. Token averages and medians use complete official run totals only. Hollow dots show partial recorded usage as lower bounds and are not included in those statistics. Each dot counts once: tokens per floor is calculated for each run before taking the mean and median, not a pooled ratio. Floor reached includes all attributed runs, regardless of outcome; a mixed-thinking tag marks a run that used more than one reasoning level with the same model. Best shows the highest floor reached in this group; ties prefer a victory. These campaigns are observational, not controlled comparisons. Unknown or mixed-model attribution and active runs are excluded. Memory follows the recorded learning policy, not the connector alone. Composition includes input, cached input, and total output when they reconcile with the run total. Uncached input includes cache writes. Reasoning and other output are shown only when the split is known; otherwise all output is shown together. Input and output bars are independently normalized to 100%, not a comparison of their volumes.