Vals Index

A composite view summarizes model performance across professional-domain evaluations.26 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
Claude Sonnet 5.5 →Anthropic	70.85% 1 report	Vals Index[2]Cost / Test: $18.70; Input / Output Cost: $2 / $10; Input / Output Tokens: 45.09B / 654.94M; Duration: 1h26m	Leads this displayed comparison. $18.70 / test1h 26m
Gemini 4 Argon →Google	69.07% 2 reports · range 68.9–69.24%	Vals Index[2][1] View 2 reports 69.24 % · Vals Index · Accuracy · Cost / Test: $13.37; Input / Output Cost: $4 / $20; Input / Output Tokens: 13.36B / 146.17M; Duration: 44m58s [2] 68.9% · Vals Index [1]	Published observations span 0.34 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty. #2 of 28 in this displayed comparison; 1.78 pp below the leader.
Claude Opus 5.5 →Anthropic	68.59% 2 reports · range 67–70.18%	Vals Index[2][1] View 2 reports 70.18 % · Vals Index · Accuracy · Cost / Test: $28.12; Input / Output Cost: $4 / $20; Input / Output Tokens: 31.96B / 499.75M; Duration: 1h25m [2] 67.0% · Vals Index [1]	Published observations span 3.18 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty. #3 of 28 in this displayed comparison; 2.26 pp below the leader.
Wonder Full →	66.79% 1 report	Vals Index[2]Cost / Test: $6.81; Input / Output Cost: $4 / $20; Input / Output Tokens: 11.05B / 96.96M; Duration: 59m19s	#4 of 28 in this displayed comparison; 4.06 pp below the leader. $6.81 / test59m 19s
Claude Fable 5.1 →Anthropic	65.8% 1 report	Vals Index[1]	#5 of 28 in this displayed comparison; 5.05 pp below the leader.
Wonder →	64.88% 1 report	Vals Index[2]Cost / Test: $10.65; Input / Output Cost: $4 / $20; Input / Output Tokens: 18.22B / 97.20M; Duration: 35m24s	#6 of 28 in this displayed comparison; 5.97 pp below the leader. $10.65 / test35m 24s
GPT-6 Astra →OpenAI	64.06% 2 reports · range 63.1–65.02%	Vals Index[2][1] View 2 reports 65.02 % · Vals Index · Accuracy · Cost / Test: $21.28; Input / Output Cost: $10 / $50; Input / Output Tokens: 4.92B / 96.77M; Duration: 33m39s [2] 63.1% · Vals Index [1]	Published observations span 1.92 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty. #7 of 28 in this displayed comparison; 6.79 pp below the leader.
GPT-6.1 Sol →OpenAI	63.94% 1 report	Vals Index[2]Cost / Test: $3.61; Input / Output Cost: $2 / $10; Input / Output Tokens: 5.49B / 103.20M; Duration: 44m24s	#8 of 28 in this displayed comparison; 6.91 pp below the leader. $3.61 / test44m 24s
Kimi K3.1 Pro →Kimi	62.52% 1 report	Vals Index[2]Cost / Test: $6.44; Input / Output Cost: $2 / $10; Input / Output Tokens: 12.82B / 188.00M; Duration: 1h04m	#9 of 28 in this displayed comparison; 8.33 pp below the leader. $6.44 / test1h 04m
DeepSeek V4.1 Flash 1017 →DeepSeek	61.26% 1 report	Vals Index[2]Cost / Test: $0.66; Input / Output Cost: $0.3 / $1.2; Input / Output Tokens: 27.11B / 324.13M; Duration: 39m55s	#10 of 28 in this displayed comparison; 9.59 pp below the leader. $0.66 / test39m 55s
GLM 5.3 →Z.ai	59.73% 1 report	Vals Index[2]Cost / Test: $8.52; Input / Output Cost: $1.4 / $4.4; Input / Output Tokens: 15.04B / 216.34M; Duration: 1h04m	#11 of 28 in this displayed comparison; 11.12 pp below the leader. $8.52 / test1h 04m
Qwen 3.8 Max 0902 →Qwen	59.05% 1 report	Vals Index[2]Cost / Test: $8.73; Input / Output Cost: $2 / $6; Input / Output Tokens: 16.16B / 266.18M; Duration: 1h46m	#12 of 28 in this displayed comparison; 11.8 pp below the leader. $8.73 / test1h 46m
Claude Haiku 5.5 →Anthropic	58.71% 1 report	Vals Index[2]Cost / Test: $3.17; Input / Output Cost: $0.1 / $0.5; Input / Output Tokens: 23.07B / 765.85M; Duration: 49m48s	#13 of 28 in this displayed comparison; 12.14 pp below the leader. $3.17 / test49m 48s
Grok 4.7 →xAI	58.56% 1 report	Vals Index[2]Cost / Test: $11.63; Input / Output Cost: $2 / $6; Input / Output Tokens: 7.30B / 112.66M; Duration: 33m34s	#14 of 28 in this displayed comparison; 12.29 pp below the leader. $11.63 / test33m 34s
MiMo V2.6 Pro →Xiaomi	57.81% 1 report	Vals Index[2]Cost / Test: $0.43; Input / Output Cost: $0.435 / $0.87; Input / Output Tokens: 9.59B / 116.02M; Duration: 1h13m	#15 of 28 in this displayed comparison; 13.04 pp below the leader. $0.43 / test1h 13m
Gemini 3.8 Flash →Google	55.95% 1 report	Vals Index[2]Cost / Test: $6.06; Input / Output Cost: $1.5 / $7.5; Input / Output Tokens: 15.37B / 117.53M; Duration: 35m58s	#16 of 28 in this displayed comparison; 14.9 pp below the leader. $6.06 / test35m 58s
MiMo V2.6 Flash →Xiaomi	54.83% 1 report	Vals Index[2]Cost / Test: $0.22; Input / Output Cost: $0.14 / $0.28; Input / Output Tokens: 14.24B / 130.89M; Duration: 1h07m	#17 of 28 in this displayed comparison; 16.02 pp below the leader. $0.22 / test1h 07m
MiniMax M3.1 Flash →MiniMax	53.78% 1 report	Vals Index[2]Cost / Test: $0.66; Input / Output Cost: $0.15 / $0.6; Input / Output Tokens: 18.18B / 306.41M; Duration: 43m18s	#18 of 28 in this displayed comparison; 17.07 pp below the leader. $0.66 / test43m 18s
Hy4 Preview →	53.21% 1 report	Vals Index[2]Cost / Test: $1.45; Input / Output Cost: $0.834 / $2.501; Input / Output Tokens: 10.73B / 186.49M; Duration: 59m23s	#19 of 28 in this displayed comparison; 17.64 pp below the leader. $1.45 / test59m 23s
DeepSeek V4 Pro 0813 →DeepSeek	52.97% 1 report	Vals Index[2]Cost / Test: $2.50; Input / Output Cost: $1.32 / $3.96; Input / Output Tokens: 22.17B / 154.91M; Duration: 58m03s	#20 of 28 in this displayed comparison; 17.88 pp below the leader. $2.50 / test58m 03s
Kimi K3 →Kimi	52.95% 1 report	Vals Index[2]Cost / Test: $7.17; Input / Output Cost: $3 / $15; Input / Output Tokens: 8.95B / 88.48M; Duration: 1h15m	#21 of 28 in this displayed comparison; 17.9 pp below the leader. $7.17 / test1h 15m
GPT-6 Luna →OpenAI	52.32% 1 report	Vals Index[2]Cost / Test: $0.45; Input / Output Cost: $0.1 / $0.5; Input / Output Tokens: 10.05B / 202.17M; Duration: 32m23s	#22 of 28 in this displayed comparison; 18.53 pp below the leader. $0.45 / test32m 23s
GLM 5.3 Flash →Z.ai	50.86% 1 report	Vals Index[2]Cost / Test: $1.13; Input / Output Cost: $0.075 / $0.25; Input / Output Tokens: 31.48B / 178.94M; Duration: 1h04m	#23 of 28 in this displayed comparison; 19.99 pp below the leader. $1.13 / test1h 04m
Muse Spark Open →Meta	48.57% 1 report	Vals Index[2]Cost / Test: $4.72; Input / Output Cost: $1.25 / $4.25; Input / Output Tokens: 10.32B / 61.83M; Duration: 1h22m	#24 of 28 in this displayed comparison; 22.28 pp below the leader. $4.72 / test1h 22m
Mercury 3 →	43.66% 1 report	Vals Index[2]Cost / Test: $0.48; Input / Output Cost: $0.1 / $0.35; Input / Output Tokens: 10.78B / 108.10M; Duration: 22m06s	#25 of 28 in this displayed comparison; 27.19 pp below the leader. $0.48 / test22m 06s
Beam 1.0 →	43.42% 1 report	Vals Index[2]Cost / Test: N/A; Input / Output Cost: N/A; Input / Output Tokens: 4.26B / 44.44M; Duration: 7m27s	#26 of 28 in this displayed comparison; 27.43 pp below the leader. 7m 27s
Inkling →	28.96% 1 report	Vals Index[2]Cost / Test: $1.85; Input / Output Cost: $1 / $4.05; Input / Output Tokens: 5.13B / 72.95M; Duration: 17m13s	#27 of 28 in this displayed comparison; 41.89 pp below the leader. $1.85 / test17m 13s
Inkling Small →	27.59% 1 report	Vals Index[2]Cost / Test: $0.40; Input / Output Cost: $0.3 / $1.2; Input / Output Tokens: 3.50B / 65.03M; Duration: 16m26s	#28 of 28 in this displayed comparison; 43.26 pp below the leader. $0.40 / test16m 26s

Sources

Gemini 4 Argon Model Card ↗
Google DeepMind · 2026
Dates & evaluation details
Vals Index · 2026-10-26
A transcription of the October model card, including its external leaderboard snapshots. These are not uniformly matched runs. Effort, tools, hardware, frame budgets and task subsets differ. GraphWalks reports F1; OSWorld uses the offline subset and partial credit. Harvey refers to the archived Vals v1 protocol, not the remediated v1.1 or AA variant. Consult the linked methodology.
Vals Index — evaluation methodology and results ↗
Vals AI · 2026
Dates & evaluation details
Vals Index · Accuracy · 2026-10-26
Public source leaderboard captured 26 October 2026. Model names, configurations, native score units and source-reported uncertainty are retained. Version and task split follow this specific Vals methodology page. These are published observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

Score vs. cost per test

Vals AI · 26 observations

Vals Index · Accuracy · higher is better

Costs are Vals AI’s published Cost / Test in USD, paired with the score from the same result. They reflect that evaluation’s workload and configuration; they are not token prices or cost per successful task. Missing costs are omitted. Explore all reported results ↓

The dashed line traces the best observed score as reported cost increases. It describes this selection of reports, with their original configurations. It does not estimate unmeasured models or costs.

View plotted values and sources (26)

Vals Index · Vals Index · Accuracy. Published measurements in their original score units.
Model	Score	Cost per test (USD)	Configuration & source
Claude Sonnet 5.5 →	70.85 %	$18.70	Cost / Test: $18.70; Input / Output Cost: $2 / $10; Input / Output Tokens: 45.09B / 654.94M; Duration: 1h26m Vals AI · Vals Index · Accuracy ↗
Claude Opus 5.5 →	70.18 %	$28.12	Cost / Test: $28.12; Input / Output Cost: $4 / $20; Input / Output Tokens: 31.96B / 499.75M; Duration: 1h25m Vals AI · Vals Index · Accuracy ↗
Gemini 4 Argon →	69.24 %	$13.37	Cost / Test: $13.37; Input / Output Cost: $4 / $20; Input / Output Tokens: 13.36B / 146.17M; Duration: 44m58s Vals AI · Vals Index · Accuracy ↗
Wonder Full →	66.79 %	$6.81	Cost / Test: $6.81; Input / Output Cost: $4 / $20; Input / Output Tokens: 11.05B / 96.96M; Duration: 59m19s Vals AI · Vals Index · Accuracy ↗
GPT-6 Astra →	65.02 %	$21.28	Cost / Test: $21.28; Input / Output Cost: $10 / $50; Input / Output Tokens: 4.92B / 96.77M; Duration: 33m39s Vals AI · Vals Index · Accuracy ↗
Wonder →	64.88 %	$10.65	Cost / Test: $10.65; Input / Output Cost: $4 / $20; Input / Output Tokens: 18.22B / 97.20M; Duration: 35m24s Vals AI · Vals Index · Accuracy ↗
GPT-6.1 Sol →	63.94 %	$3.61	Cost / Test: $3.61; Input / Output Cost: $2 / $10; Input / Output Tokens: 5.49B / 103.20M; Duration: 44m24s Vals AI · Vals Index · Accuracy ↗
Kimi K3.1 Pro →	62.52 %	$6.44	Cost / Test: $6.44; Input / Output Cost: $2 / $10; Input / Output Tokens: 12.82B / 188.00M; Duration: 1h04m Vals AI · Vals Index · Accuracy ↗
DeepSeek V4.1 Flash 1017 →	61.26 %	$0.66	Cost / Test: $0.66; Input / Output Cost: $0.3 / $1.2; Input / Output Tokens: 27.11B / 324.13M; Duration: 39m55s Vals AI · Vals Index · Accuracy ↗
GLM 5.3 →	59.73 %	$8.52	Cost / Test: $8.52; Input / Output Cost: $1.4 / $4.4; Input / Output Tokens: 15.04B / 216.34M; Duration: 1h04m Vals AI · Vals Index · Accuracy ↗
Qwen 3.8 Max 0902 →	59.05 %	$8.73	Cost / Test: $8.73; Input / Output Cost: $2 / $6; Input / Output Tokens: 16.16B / 266.18M; Duration: 1h46m Vals AI · Vals Index · Accuracy ↗
Claude Haiku 5.5 →	58.71 %	$3.17	Cost / Test: $3.17; Input / Output Cost: $0.1 / $0.5; Input / Output Tokens: 23.07B / 765.85M; Duration: 49m48s Vals AI · Vals Index · Accuracy ↗
Grok 4.7 →	58.56 %	$11.63	Cost / Test: $11.63; Input / Output Cost: $2 / $6; Input / Output Tokens: 7.30B / 112.66M; Duration: 33m34s Vals AI · Vals Index · Accuracy ↗
MiMo V2.6 Pro →	57.81 %	$0.43	Cost / Test: $0.43; Input / Output Cost: $0.435 / $0.87; Input / Output Tokens: 9.59B / 116.02M; Duration: 1h13m Vals AI · Vals Index · Accuracy ↗
Gemini 3.8 Flash →	55.95 %	$6.06	Cost / Test: $6.06; Input / Output Cost: $1.5 / $7.5; Input / Output Tokens: 15.37B / 117.53M; Duration: 35m58s Vals AI · Vals Index · Accuracy ↗
MiMo V2.6 Flash →	54.83 %	$0.22	Cost / Test: $0.22; Input / Output Cost: $0.14 / $0.28; Input / Output Tokens: 14.24B / 130.89M; Duration: 1h07m Vals AI · Vals Index · Accuracy ↗
MiniMax-M3.1-Flash →	53.78 %	$0.66	Cost / Test: $0.66; Input / Output Cost: $0.15 / $0.6; Input / Output Tokens: 18.18B / 306.41M; Duration: 43m18s Vals AI · Vals Index · Accuracy ↗
Hy4 Preview →	53.21 %	$1.45	Cost / Test: $1.45; Input / Output Cost: $0.834 / $2.501; Input / Output Tokens: 10.73B / 186.49M; Duration: 59m23s Vals AI · Vals Index · Accuracy ↗
DeepSeek V4 Pro 0813 →	52.97 %	$2.50	Cost / Test: $2.50; Input / Output Cost: $1.32 / $3.96; Input / Output Tokens: 22.17B / 154.91M; Duration: 58m03s Vals AI · Vals Index · Accuracy ↗
Kimi K3 →	52.95 %	$7.17	Cost / Test: $7.17; Input / Output Cost: $3 / $15; Input / Output Tokens: 8.95B / 88.48M; Duration: 1h15m Vals AI · Vals Index · Accuracy ↗
GPT-6 Luna →	52.32 %	$0.45	Cost / Test: $0.45; Input / Output Cost: $0.1 / $0.5; Input / Output Tokens: 10.05B / 202.17M; Duration: 32m23s Vals AI · Vals Index · Accuracy ↗
GLM 5.3 Flash →	50.86 %	$1.13	Cost / Test: $1.13; Input / Output Cost: $0.075 / $0.25; Input / Output Tokens: 31.48B / 178.94M; Duration: 1h04m Vals AI · Vals Index · Accuracy ↗
Muse Spark Open →	48.57 %	$4.72	Cost / Test: $4.72; Input / Output Cost: $1.25 / $4.25; Input / Output Tokens: 10.32B / 61.83M; Duration: 1h22m Vals AI · Vals Index · Accuracy ↗
Mercury 3 →	43.66 %	$0.48	Cost / Test: $0.48; Input / Output Cost: $0.1 / $0.35; Input / Output Tokens: 10.78B / 108.10M; Duration: 22m06s Vals AI · Vals Index · Accuracy ↗
Inkling →	28.96 %	$1.85	Cost / Test: $1.85; Input / Output Cost: $1 / $4.05; Input / Output Tokens: 5.13B / 72.95M; Duration: 17m13s Vals AI · Vals Index · Accuracy ↗
Inkling Small →	27.59 %	$0.40	Cost / Test: $0.40; Input / Output Cost: $0.3 / $1.2; Input / Output Tokens: 3.50B / 65.03M; Duration: 16m26s Vals AI · Vals Index · Accuracy ↗

Reported in recent model launches

These model reports include this benchmark. Their harnesses, task subsets, scoring policies, and resource budgets can differ; follow each report’s evaluation notes before joining the numbers.

Google · Gemini 4 Argon →

What it measures

A composite view summarizes model performance across professional-domain evaluations.

How to read the result

Inspect the component benchmarks and their weights before interpreting the aggregate as a broad work-capability measure.

Limits and comparison traps

Changes in composition, model configuration, or source snapshots change the meaning of the index. A composite does not replace its underlying results.

Published evaluation

The linked publication documents the task and scoring procedure. Model results retain the source’s evaluation setting, native score units and available uncertainty.

Where this fits in an evaluation suite

Read accuracy together with error and abstention rates. Domain averages can hide language gaps, specialist failures, and poorly calibrated confidence.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Pin the benchmark release, task split, model configuration, tools, sampling budget, and evaluation harness described by the primary source.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

Vals Index — evaluation methodology and results ↗
Vals AI · 2026
Gemini 4 Argon Model Card ↗
Google DeepMind · 2026

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All knowledge & truth benchmarks →