BenchLeader

Token efficiency: quality per dollar of real work

BenchLeader Index divided by what an average task of Artificial Analysis' Intelligence Index cost on that configuration. Unlike price per token this counts the tokens a model actually burns, so a model that thinks at length scores lower than its sticker price suggests.

As of 15 Nov 2026, BenchLeader ranks GPT-6 Luna first on the Token efficiency: quality per dollar of real work board at 8945, ahead of Claude Haiku 5.5 (1090) and Granite 4.2 8B (1016).

120 models · data as of 15 Nov 2026

Size

Quality against what it costs to get there

5560657075$0.2$0.5$1$2$5$10$20Cost per Intelligence Index task, USD (log scale)BenchLeader Index v1.4Claude Fable 5.2Gemini 4 CarbonClaude Opus 5.5GPT-6 AstraGPT-6.1 SolClaude Fable 5.1

Hover or tap a model to pick it out and see its other effort settings; click to pin it (Esc clears). The eye hides a model.

Cost per task is what an average task of the Intelligence Index cost at list price, averaged over Artificial Analysis’ whole run of that configuration (source: Artificial Analysis). It counts the tokens a model really writes, so a model that thinks at length pays for it here. Each dot is a model at its best reasoning-effort setting (highest Index); pick a model out to see its other settings, joined lowest effort first; up and to the left is better.

How far the effort dial moves each model

One bar per model, from its lowest to its highest figure across its reasoning-effort settings, a mark at each setting. The same models as the chart above.

Sort

Tokens per task

output tokens, reasoning included
0100k200kGPT-6 Astra9.1k – 52klowmaxGPT-6.1 Sol5.8k – 68klowmaxClaude Fable 5.138k – 118klowmaxGemini 4 Carbon86k – 123klowhighClaude Opus 5.519k – 173klowmaxClaude Fable 5.236k – 274klowmax

Output tokens, full Index run

every task of the run, as AA totals it
0100M200MGPT-6 Astra10.3M – 54.3MlowmaxGPT-6.1 Sol8.2M – 63.7MlowmaxClaude Fable 5.136.3M – 163MlowmaxGemini 4 Carbon83.9M – 205MlowhighClaude Opus 5.521.1M – 218MlowmaxClaude Fable 5.234.9M – 275Mlowmax

Artificial Analysis’ own total for the whole run: nine evaluations and over 5,000 tasks, several run three or five times. Its per-task figures are averages weighted by each evaluation’s share of the index, so they do not multiply back to this.

Cost per task

USD at list price
$0$10$20GPT-6.1 Sol$0.161 – $1.27Gemini 4 Carbon$4.68 – $5.48GPT-6 Astra$1.42 – $6.24lowmaxClaude Opus 5.5$0.976 – $10lowmaxClaude Fable 5.1$4.47 – $13lowmaxClaude Fable 5.2$3.10 – $27lowmax

Cost of a full Index run

USD at list price, as AA totals it
$0$10k$20kGPT-6.1 Sol$238 – $1.1kGemini 4 Carbon$4k – $5.3kGPT-6 Astra$1.7k – $5.7klowmaxClaude Opus 5.5$980 – $8.9klowmaxClaude Fable 5.1$3.9k – $13.3klowmaxClaude Fable 5.2$2.9k – $22.2klowmax

Artificial Analysis’ own total for the whole run: nine evaluations and over 5,000 tasks, several run three or five times. Its per-task figures are averages weighted by each evaluation’s share of the index, so they do not multiply back to this.

Time per task

per Index task
010 min20 min30 min40 minGPT-6 Astra3.5 min – 21.8 minlowmaxClaude Fable 5.17.6 min – 22.3 minlowmaxGPT-6.1 Sol2 min – 22.9 minlowmaxClaude Opus 5.53.1 min – 25.4 minlowmaxGemini 4 Carbon28.1 min – 33.8 minlowhighClaude Fable 5.25.6 min – 44.5 minlowmax

Cost, tokens and time per task, and the full-run totals, are from Artificial Analysis’ Intelligence Index runs, shown with credit and not offered for reuse. The Index is BenchLeader’s.

58 of 58
#
1GPT-6 Lunalow894545.8$0.0051–
2Claude Haiku 5.5low109047.7$0.044–
3Granite 4.2 8B101641.1$0.040Small
4Mistral Small 4thinking101540.5$0.040Medium
5GPT-6.1 Terralow79356.6$0.071–
6MiMo-V2.6-Flash42654.8$0.129Large
7NewMuse Emberxhigh42249.1$0.116–
8GPT-6.1 Sollow37260.0$0.161–
9Gemini 3.1 Flash Lite27052.4$0.194–
10Muse Glimmerhigh26146.5$0.178Small
11Inkling Small25652.0$0.203Large
12MiMo-V2.6-Pro22156.8$0.257Large
13GPT-5.6 Lunamax17056.2$0.331–
14Beam 1.016145.9$0.284Large
15Nemotron 3.5 Lightning13241.4$0.314Small
16gpt-oss-120bhigh13145.3$0.345Medium
17DeepSeek V4.1 Flash 1017high11251.2$0.456Large
18MiniMax M3.1 Flash10953.5$0.491Large
19Claude Sonnet 5.5low99.250.7$0.511–
20GLM 5.3 Flash96.956.5$0.583Large
21Qwen3.8-Flash-Next94.753.9$0.569Large
22NewAgnes 2.5 Pro Theta88.450.8$0.575–
23Claude Haiku 4.5thinking85.245.3$0.532–
24Nemotron 3.5 Super VL82.245.8$0.557Medium
25Gemini 3.5 Flash Lite74.251.4$0.693–
26DeepSeek V4 Prohigh66.056.0$0.849Large
27Apodex 1.162.848.0$0.764–
28Claude Opus 5.5low60.458.9$0.976–
29Hy455.056.7$1.03–
30Solar Pro 451.248.9$0.956–
31Inkling 2 Small49.157.9$1.18Large
32Mistral Medium 3.544.846.6$1.04Medium
33MiniMax M343.756.1$1.29Large
34GPT-6 Astralow43.562.0$1.42–
35Inklingxhigh37.050.0$1.35Large
36NewGemini 4 Hydrogenhigh33.863.4$1.87–
37Mistral Large 4medium32.545.9$1.41Large
38GLM 5.3low30.047.3$1.57Large
39Muse Spark Openxhigh26.553.5$2.02Large
40Muse Spark 1.3max25.861.5$2.39–
41Qwen3.8 27Blow25.749.0$1.91Small
42GPT-6 Solmax25.462.6$2.47–
43Step 5 Preview23.154.2$2.35Large
44Kimi K3low22.653.6$2.37Large
45Nemotron 3 Ultra 550B A55Bthinking22.249.1$2.21Large
46NewClaude Fable 5.2low21.165.3$3.10–
47Grok 4.7low19.055.4$2.92–
48Kimi K3.1 Promax16.761.3$3.68Large
49GPT-5.6 Terramax16.260.4$3.73–
50Ling 3.1 Flash15.950.8$3.19Large
51Gemini 4 Argonmedium14.361.6$4.31–
52Gemini 4 Carbonmedium14.165.8$4.68–
53Claude Fable 5.1low13.258.8$4.47–
54Muse Wonderhigh12.356.2$4.56–
55Grok 4.6xhigh12.257.3$4.68–
56Gemini 3.8 Flashmedium9.660.2$6.26–
57Qwen3.8 Maxmax5.958.7$10.02–
58Claude Opus 5max5.564.8$11.82–

Cite as: BenchLeader, “Token efficiency: quality per dollar of real work”, data as of 15 Nov 2026, https://www.benchleader.com/leaderboards/efficiency.