ai benchmarks

Real-world AI performance metrics

Gauntlet Composite updated

Gauntlet Science updated

Gauntlet Ship updated

Gauntlet Teach updated

Gauntlet Judgment updated

Artificial Analysis data updated

Every change Artificial Analysis made to its scores and prices, newest first. Click a model to see its past scores.

1 price change

Price changes 1

  • single settingCost per task$0.21 → $0.30+$0.095

1 new release

New releases 1

  • peak 26.5one setting, $0.21 per task

1 new release

New releases 1

  • peak 38.9one setting, $0.84 per task

6 new releases · 1 price change

New releases 6

  • peak 21.8one setting, $0.019 per task
  • peak 10.8one setting, $0.039 per task
  • peak 9.3one setting, $0.038 per task
  • peak 8.9one setting, $0.0082 per task
  • peak 7.0one setting, $0.11 per task
  • peak 6.5one setting, $0.11 per task

Price changes 1

  • single settingCost per task$1.00 → $1.00−$0.0020

1 price change

Price changes 1

  • highCost per task$0.057 → $0.053−$0.0032

1 price change

Price changes 1

  • single settingCost per task$1.13 → $1.13+$0.0023

1 price change

Price changes 1

  • single settingCost per task$0.25 → $0.28+$0.030

1 new release

New releases 1

  • peak 52.4one setting, $4.52 per task

1 removed release · 3 price changes

Price changes 3

  • highCost per task$0.12 → $0.10−$0.021
  • DeepSeek V4 Flash 0420maxCost per task$0.11 → $0.089−$0.018
  • single settingCost per task$0.077 → $0.066−$0.011

Removed releases 1

  • was 12.7 peakhad one setting

2 price changes

Mistral Small 3.1’s cost per Intelligence Index task fell from $0.035 to $0.0255. OpenAI’s gpt-oss-20b at high reasoning effort also became cheaper, from $0.0131 to $0.0125.

AI-written summary of these changes · gpt-6-luna

Price changes 2

  • single settingCost per task$0.035 → $0.025−$0.0095
  • highCost per task$0.013 → $0.013−$0.0006

2 price changes

The cost per Intelligence Index task rose for both models: Qwen3 Coder Next went from $0.5518 to $0.7805, while gpt-oss-20b at high reasoning effort went from $0.0125 to $0.0131. Qwen3 Coder Next had the larger increase, both in dollars and proportionally.

AI-written summary of these changes · gpt-6-luna

Price changes 2

  • single settingCost per task$0.55 → $0.78+$0.23
  • highCost per task$0.013 → $0.013+$0.0006

3 price changes

All three listed cost per task figures increased slightly. Ling 3.1 Flash rose from $0.9857 to $1.002, Solar Mini 4 from $0.3634 to $0.3656, and MiMo-V2.6-Flash from $0.0623 to $0.0624.

AI-written summary of these changes · gpt-6-luna

Price changes 3

  • single settingCost per task$0.99 → $1.00+$0.016
  • single settingCost per task$0.36 → $0.37+$0.0022
  • single settingCost per task$0.062 → $0.062+$0.0001

1 price change

LongCat 2.0’s cost per Intelligence Index task rose from $0.059 to $0.1462. This was the only listed change.

AI-written summary of these changes · gpt-6-luna

Price changes 1

  • single settingCost per task$0.059 → $0.15+$0.087

1 score change

Claude Haiku 5.5’s Intelligence Index score at low reasoning effort rose from 29.4 to 29.5. This is the only change in the update.

AI-written summary of these changes · gpt-6-luna

Score changes 1

  • lowIntelligence29.4 → 29.5+0.1

1 new release

IBM added Granite 4.2 30B with an Intelligence Index score of 12.7. Its cost is $0.072 per task.

AI-written summary of these changes · gpt-6-luna

New releases 1

  • peak 12.7one setting, $0.072 per task

1 price change

StepFun’s Step 5 Preview default cost per Intelligence Index task rose from $0.7175 to $1.026, an increase of $0.3085 or about 43%.

AI-written summary of these changes · gpt-5.6-terra

Price changes 1

  • single settingCost per task$0.72 → $1.03+$0.31

5 price changes

Claude Sonnet 5.5’s Intelligence task cost fell at every listed reasoning effort. Max dropped from $7.667 to $5.461, xhigh from $2.746 to $2.012, and high from $1.123 to $0.8844. Medium and low also declined, from $0.5894 to $0.4833 and from $0.4170 to $0.3452.

AI-written summary of these changes · gpt-5.6-terra

Price changes 5

  • lowCost per task$0.42 → $0.35−$0.072
  • Claude Sonnet 5.5mediumCost per task$0.59 → $0.48−$0.11
  • Claude Sonnet 5.5highCost per task$1.12 → $0.88−$0.24
  • Claude Sonnet 5.5xhighCost per task$2.75 → $2.01−$0.73
  • Claude Sonnet 5.5maxCost per task$7.67 → $5.46−$2.21

1 new release

Anthropic’s Claude Haiku 5.5 was added with an Intelligence Index score of 43.4. It is available at five effort levels, with Intelligence task costs ranging from $0.025 to $0.21.

AI-written summary of these changes · gpt-5.6-terra

New releases 1

  • peak 43.45 effort levels, $0.025–$0.21 per task

1 price change

DeepSeek V4 Flash 0420’s high-effort Intelligence task cost fell from $0.1271 to $0.1231, a decrease of $0.0040 per task.

AI-written summary of these changes · gpt-5.6-terra

Price changes 1

  • highCost per task$0.13 → $0.12−$0.0040

1 new release

Mistral Large 4 Preview entered the Intelligence Index at 38.4, with its listed setting costing $1.13 per task.

AI-written summary of these changes · gpt-5.6-terra

New releases 1

  • peak 38.4one setting, $1.13 per task
Load older