Get started

Epoch Capabilities Index

One number per model, fit by Epoch over the benchmarks that model has results on, with the interval around it. 258 models placed, 116 of them listed in the catalog Vitna prices.

retrieved ·CC-BY-4.0epoch.ai/eci ↗

Models placed258each on 4 benchmark results or more
Results behind it2,917counted here, not republished
Scale130 / 150anchored at Claude 3.5 Sonnet and GPT-5
Released2023 to 2026by the date each model came out

What the interval says

TOP 24 OF 258

Claude Fable 5.2 leads at 173.4, and 8 other models have intervals reaching its own, so the index does not separate them.

  1. 01Claude Fable 5.2Anthropic173.4
  2. 02GPT-6.2 AstraOpenAI172.8
  3. 03Gemini 4 CarbonGoogle DeepMind169.7
  4. 04GPT-6 AstraOpenAI169.2
  5. 05Claude Opus 5.5Anthropic168.7
  6. 06GPT-6.1 SolOpenAI168.6
  7. 07GPT-6.1 TerraOpenAI167.7
  8. 08Claude Fable 5.1Anthropic167.1
  9. 09Gemini 4 HydrogenGoogle DeepMind166.8
  10. 10Claude Sonnet 5.5Anthropic166.6
  11. 11Gemini 4 ArgonGoogle DeepMind166.5
  12. 12Claude Opus 5Anthropic164.7
  13. 13Claude Fable 5Anthropic164.2
  14. 14GPT-6 SolOpenAI163.8
  15. 15GPT-5.6 SolOpenAI163.0
  16. 16GPT-5.5 ProOpenAI162.8
  17. 17Muse WonderMeta AI161.7
  18. 18Kimi K3.1 ProMoonshot161.2
  19. 19GPT-5.6 TerraOpenAI160.8
  20. 20Claude Haiku 5.5Anthropic160.3
  21. 21GPT-5.4 ProOpenAI160.3
  22. 22GPT-5.5OpenAI160.2
  23. 23Claude Opus 4.8Anthropic159.9
  24. 24DeepSeek V4.1 Flash 1017DeepSeek159.4

The rail runs 157.1 to 177.2, the range these 24 rows cover rather than the whole index, because scaled to all 258 they would sit on top of one another. Each bar is the interval Epoch publishes and the mark is the figure; a marked row is one whose interval reaches that of the leader. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.

Vitna fits a scale of its own over the six openly licensed boards, 319 ranked runs over 46 columns, and this is the published index the ordering of that fit is checked against. The two are separate fits over different tables, so their numbers do not compare and only their orderings do.

46 of 258.

Index figures, as Epoch publishes them
#ModelOrganisationWeightsReleasedResultsIntervalIndex
2GPT-6.2 AstraOpenAIClosed weights15169.4 to 176.9172.8
4GPT-6 AstraOpenAIClosed weights26166.4 to 172.5169.2
6GPT-6.1 SolOpenAIClosed weights21165.7 to 171.9168.6
7GPT-6.1 TerraOpenAIClosed weights13164.9 to 171.0167.7
14GPT-6 SolOpenAIClosed weights20161.4 to 166.6163.8
15GPT-5.6 SolOpenAIClosed weights27160.7 to 165.6163.0
16GPT-5.5 ProOpenAIClosed weights10160.1 to 165.9162.8
19GPT-5.6 TerraOpenAIClosed weights20158.5 to 163.3160.8
21GPT-5.4 ProOpenAIClosed weights9157.9 to 163.0160.3
22GPT-5.5OpenAIClosed weights28158.2 to 162.4160.2
29GPT-5.4OpenAIClosed weights24156.0 to 160.0157.9
33GPT-5.2 ProOpenAIClosed weights6155.1 to 159.9157.3
37GPT-5.6 LunaOpenAIClosed weights21154.9 to 158.8156.7
38GPT-6 LunaOpenAIClosed weights18154.6 to 158.5156.4
52GPT-5.2OpenAIClosed weights26152.9 to 156.3154.5
65GPT-5 ProOpenAIClosed weights7149.4 to 153.3151.3
67GPT-5.1OpenAIClosed weights19149.3 to 152.6150.9
72GPT-5OpenAIClosed weights28anchor, fixed150.0
75GPT-5.4 MiniOpenAIClosed weights15147.7 to 151.1149.4
83o3-proOpenAIClosed weights7145.7 to 149.4147.5
85o3OpenAIClosed weights25145.5 to 148.6147.0
91GPT-5.4 NanoOpenAIClosed weights13144.7 to 148.0146.4
93GPT-5 miniOpenAIClosed weights17144.5 to 147.6146.0
96o4-miniOpenAIClosed weights22143.7 to 146.7145.2
119o1OpenAIClosed weights17139.5 to 142.8141.2
125o3-miniOpenAIClosed weights19138.9 to 142.3140.6
127gpt-oss-120bOpenAIOpen weights13138.3 to 142.0140.2
130GPT-5 nanoOpenAIClosed weights17137.1 to 140.9139.1
137GPT-4.5OpenAIClosed weights11134.4 to 138.4136.5
139GPT-4.1OpenAIClosed weights17134.2 to 138.0136.1
142o1-miniOpenAIClosed weights10133.0 to 137.3135.3
144gpt-oss-20bOpenAIOpen weights6132.3 to 137.3134.9
146GPT-4.1 miniOpenAIClosed weights15132.1 to 136.2134.2
150o1-previewOpenAIClosed weights8131.1 to 135.9133.6
162GPT-4o (Nov 2024)OpenAIClosed weights15126.1 to 131.2128.8
163GPT-4o (May 2024)OpenAIClosed weights8125.4 to 130.7128.1
168GPT-4o (Aug 2024)OpenAIClosed weights10124.1 to 129.9127.2
170GPT-4.1 nanoOpenAIClosed weights13123.8 to 129.5126.8
173GPT-4 Turbo (Apr 2024)OpenAIClosed weights8122.5 to 128.8125.8
176GPT-4 Turbo (Nov 2023)OpenAIClosed weights6121.5 to 127.7124.8
179GPT-4o miniOpenAIClosed weights19120.5 to 126.8123.8
182GPT-4 (Mar 2023)OpenAIClosed weights7117.6 to 125.2121.6
187GPT-4 (Jun 2023)OpenAIClosed weights11116.2 to 123.3119.9
201GPT-3.5 Turbo (Jan 2024)OpenAIClosed weights9107.8 to 117.3112.9
202GPT-3.5 Turbo (Nov 2023)OpenAIClosed weights8107.3 to 117.2112.7
210GPT-3.5 Turbo (Jun 2023)OpenAIClosed weights497.6 to 113.4106.0

46 of 258 models, highest index first. A model links to its catalog page where the pairing names one, which is true of 116; the rest are models the catalog does not list, and the record Epoch keeps of them stands on its own. Results is how many benchmark results stand behind the figure, and Epoch places nothing on fewer than 4. The two rows with no interval are the anchors the scale is defined by, whose figures are fixed rather than estimated.

Sources

Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
The index is Epoch's fit. Vitna ran none of these evaluations, computed none of these figures and adjusted none of them.