RankingEffort benchmarks

Hy3 · Reasoning

Compare models

18 measured results · 3 effort settings · 18 benchmark/harness combinations

Model evidence and sources →

Change model

Model

Current configurationSources

Compare with

Choose a second model and its effort to see both profiles and benchmark differences below.

Compare withModel and configuration

Capability profile

One configuration, seven capabilities

Comparison score · 0–100

AHy3 · Reasoning2 of 7 capabilities supported

Gaps are unknown, not zero. Hollow points are preliminary.

Agentic
A29.6Supported
Hard reasoning
A31.4Supported
Coding
A25.6Preliminary
Human pref
Unknown
Knowledge
A23.3Preliminary
Multimodal
Unknown
Long context
A45.0Preliminary

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Capability score evidence

Hy3 · Reasoning: 19.3 overall comparison score. Meets the overall evidence requirements.

5 areas measured. The available task mix affects this estimate; missing areas remain unknown.

Which sources carry the weight?

Artificial Analysis · 100.0% direct comparison weight · independent of the model provider

Which benchmark families carry the weight?

Humanity’s Last Exam · 10.7% direct comparison weight
GPQA · 10.7% direct comparison weight
Critpt · 10.6% direct comparison weight
omniscience · 10.6% direct comparison weight
Aa Lcr · 9.8% direct comparison weight
Other families (6) · 47.5% direct comparison weight

These shares describe direct comparison weight, not fractions of the final score or confidence. Source balancing does not give every family equal weight or remove correlations between overlapping sources.

How sensitive is this score?

The audit removes one source at a time and refits against the same overall reference panel. These published results use the matching dataset audited 2026-10-23.

Both ranks use the same surviving configurations for each removal. The cohort starts with 444 configurations supported in the original audit, including separate effort settings. These are diagnostic positions, not the default leaderboard ranks. Eligibility is not rechecked after removal.

What changes when a source is removed?
Removed source	Rank before → after	Rank movement	Cohort remaining / lost support	Score after (change)
Artificial Analysis	Comparison support lost; no rank		203 remaining / 241 lost	Unavailable
ARC Prize	138 → 140	Down 2	444 remaining / 0 lost	23.5 (+4.2 points)
Datacurve	138 → 138	No movement	444 remaining / 0 lost	15.9 (-3.4 points)
LiveCodeBench	138 → 139	Down 1	444 remaining / 0 lost	19.2 (-0.1 points)
Scale	138 → 138	No movement	444 remaining / 0 lost	21.4 (+2.1 points)
Terminal-Bench	138 → 138	No movement	444 remaining / 0 lost	19.2 (-0.1 points)
Cognition	138 → 141	Down 3	444 remaining / 0 lost	18.4 (-0.9 points)
LiveBench	137 → 138	Down 1	443 remaining / 1 lost	19.8 (+0.5 points)
SWE-rebench	138 → 138	No movement	444 remaining / 0 lost	19.7 (+0.4 points)
Epoch AI	136 → 137	Down 1	431 remaining / 13 lost	19.6 (+0.3 points)
Vals AI	138 → 137	Up 1	443 remaining / 1 lost	18.6 (-0.7 points)
Provider reports · Anthropic	138 → 138	No movement	444 remaining / 0 lost	19.3 (-0.0 points)
Provider reports · Moonshot AI	138 → 139	Down 1	444 remaining / 0 lost	18.8 (-0.5 points)
Provider reports · Meta	138 → 138	No movement	444 remaining / 0 lost	19.3 (-0.0 points)
Provider reports · Mistral	138 → 138	No movement	444 remaining / 0 lost	19.3 (+0.0 points)
Provider reports · OpenAI	138 → 139	Down 1	444 remaining / 0 lost	19.4 (+0.1 points)
All provider reports	138 → 139	Down 1	444 remaining / 0 lost	18.9 (-0.4 points)

Lost support means a configuration no longer has a comparison path to the complete reference panel. Its rank is unavailable. A surviving path alone does not establish leaderboard eligibility.

Overlapping sources may remain correlated even with source balancing. Leave-one-source-out movement is a sensitivity diagnostic, not a spread of source-specific ranks or a confidence measure. Scoring methodology · Dated configuration audit

Every benchmark family

Removed evidence	Recomputed score	Change
arc-agi	22.9	+3.6
deepswe	15.9	-3.4
mmmu-pro	20.5	+1.2
gpqa	18.2	-1.1
aa-lcr	18.3	-1.0
frontiercode	18.4	-0.9
hle	20.1	+0.8
epoch-game-puzzles	18.6	-0.7
automationbench	20.0	+0.7
simpleqa	20.0	+0.7
mmlu-pro	19.7	+0.4
ifbench	18.9	-0.4
terminal-bench	19.7	+0.4
swe-rebench	19.7	+0.4
omniscience	19.7	+0.4
tau-bench	18.9	-0.4
livebench-instructions	19.6	+0.3
multi-swe-bench	19.0	-0.3
gdp-pdf	19.6	+0.3
apex-agents	19.6	+0.3
livebench-language	19.5	+0.2
harvey	19.1	-0.2
briefcase	19.1	-0.2
arc-agi-3	19.1	-0.2
terminal-bench-science	19.5	+0.2
swe-atlas-qna	19.1	-0.2
enigma-eval	19.1	-0.2
livebench-coding	19.1	-0.2
analyst-agent	19.5	+0.2
gdpval	19.5	+0.2
critpt	19.5	+0.1
gmmlu	19.4	+0.1
itbench	19.2	-0.1
livebench-reasoning	19.2	-0.1
vibe-code	19.2	-0.1
swe-atlas-test-writing	19.2	-0.1
vals-legal-research	19.2	-0.1
vals-finance-agent	19.2	-0.1
vals-code-migration	19.3	-0.1
vals-tax-agent	19.3	-0.0
aime	19.3	-0.0
math500	19.3	+0.0
vals-excel-modeling	19.3	-0.0
livecodebench	19.3	-0.0
frontiermath	19.3	+0.0
livebench-math	19.3	+0.0
swe-atlas-refactoring	19.3	-0.0
enterprise-ops	19.3	-0.0
swe-bench-pro	19.3	+0.0
livebench-data	19.3	-0.0

What evidence is missing from the fit?

Across all collected settings for this model: 25 observations, 12 contributing, 0 matched without graph weight and 13 excluded. Counts do not establish rank eligibility.

1 · Benchmark family or source protocol has not been reviewed
1 · Original SciCode grading defects are under review; retained as supporting evidence, excluded from aggregate and capability scores. Corrected SciCode-Verified requires a separate protocol review. Sources: https://arxiv.org/abs/2608.04975 and https://artificialanalysis.ai/evaluations/scicode
3 · Evaluator has not been reviewed for independent admission
1 · Effort is not identified
5 · Supporting evidence outside the reviewed capability core
2 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately

Inspect every observation and exclusion →

Benchmark results

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Benchmark / source	High	Reasoning	Unspecified
AA-Briefcase EloArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details AA-Briefcase v1.1 is an agentic knowledge work benchmark developed by Artificial Analysis. AA-Briefcase Elo is a combined metric that aggregates rubric pass rate, analytical quality Elo and presentation Elo · Higher is better · Evaluation results measured independently by Artificial Analysis Unit: elo. Source collected 2026-10-18.	Not reported	946.66 Elo Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	79% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
AutomationBench-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of task objectives completed with no guardrail violations · Higher is better · Benchmark developed by Zapier · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	18.41% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
CritPtArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally. CritPt is under review by its authors following external feedback · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	4.86% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
GDP.pdf: All-pass RateArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of attempts where every atomic criterion passed · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	11.2% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
GDPval-AA v2.1 LeaderboardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Elo rating for performance on real-world work tasks · Anchored to DeepSeek V4.1 Flash (max) at 1600 · Higher is better · Evaluation results measured independently by Artificial Analysis Unit: elo. Source collected 2026-10-18.	Not reported	1,066.59 Elo Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	89.7% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	33.46% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
AA-Omniscience IndexArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis Unit: index. Source collected 2026-10-18.	Not reported	-18.47 Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
AA-Omniscience AccuracyArtificial Analysis · 6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader Benchmark details AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	31.97% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
SciCodeArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark curated by scientists across 16 disciplines. SciCode is under review following an independent audit of the dataset · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	48.61% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
𝜏³-BankingArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by Sierra Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	22.89% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
Terminal-Bench v2.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by the Laude Institute and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	64.42% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
Terminal-Bench v4.0Artificial Analysis · mini-SWE-agent v2.4.6; 66 tasks; pass@1 averaged over 3 repeats; 500 steps; task timeout up to 8 hours Benchmark details Benchmark developed by the Laude Institute, Stanford, and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	0.51% Reported settings Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified	Not reported
DAYJOB HealthcareSurge AI · OpenHands SDK 1.43.1 inside Harbor 0.22.0. Judge is Claude Opus 4.8. Score is the board's strict pass@1 percent. Benchmark details DAYJOB Healthcare strict pass@1 percent on the public Surge board. tau = 1, five attempts per task, 50 tasks in the paper and the Hugging Face dataset card. The page meta description says 65 assignments. That count is not on the scoreboard and is not used as n. An attempt passes only when every binary rubric criterion is met. Finance is a separate benchmark. Display only. Not an average and not GDPval-AA. Unit: percent. Source collected 2026-10-18.	0.4% Reported settings Hy3; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.	Not reported	Not reported
DAYJOB FinanceSurge AI · OpenHands SDK 1.43.1 inside Harbor 0.22.0. Judge is Claude Opus 4.8. Score is the board's strict pass@1 percent. Benchmark details DAYJOB Finance strict pass@1 percent on the public Surge board. tau = 1, five attempts per task, 80 tasks in the paper and the Hugging Face dataset card. An attempt passes only when every binary rubric criterion is met. Healthcare is a separate benchmark. Display only. Not an average and not GDPval-AA. Unit: percent. Source collected 2026-10-18.	0.3% Reported settings Hy3; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.	Not reported	Not reported
Toolathlon-VerifiedHKUST NLP · Shared Default agent. One harness for the board so Pass@1 rank matches the published table. Effort stays on the result. Benchmark details Official Toolathlon-Verified board, released June 30, 2026. Same 108 tasks, new score series, not comparable with earlier Toolathlon. Headline is printed Pass@1 on the shared Default agent. Display only. The ± is recorded in the configuration and is not the score. Historical rows are excluded. Unit: percent. Source collected 2026-10-18.	56.8% Reported settings Hy3 (high); Open-Weights; Default agent. Pass@1 56.8 ± 3.7; 108 tasks; evaluation 2026-07-30. Pass@3 72.2; Pass^3 40.7; turns 20.5; tool calls 37.9. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.	Not reported	Not reported
LMArena Agent ArenaLMArena · Arena Agent Mode. The board does not name a per-model scaffold. Benchmark details Aggregate Agent Arena IPS net improvement, stored as the board's printed percentage points. Arena Agent Mode does not name a per-model scaffold. Display only. Excluded from weighted Overall. Unit: index. Source collected 2026-10-18.	Not reported	Not reported	-5.38 Reported settings Hy3; Arena Agent Mode; IPS net improvement -5.38 points ±1.21. rank 41; observations 1171568; sessions 29677; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.

What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 8,916 results across 259 models · Collected 2026-10-18

Hy3 · Reasoning

Compare with

One configuration, seven capabilities

Agentic

Hard reasoning

Coding

Human pref

Knowledge

Multimodal

Long context

Capability score evidence

Benchmark results