RankingEffort benchmarks

DeepSeek V4-Pro 0813 · Max

Compare models

55 measured results · 6 effort settings · 41 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

ADeepSeek V4-Pro 0813 · Max4 of 7 capabilities supported

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

Agentic
A48.3Supported
Hard reasoning
A49.5Supported
Coding
A48.2Supported
Human pref
Unknown
Knowledge
A38.0Supported
Multimodal
Unknown
Long context
A54.8Preliminary

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

DeepSeek V4-Pro 0813 · Max: 44.4 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 · 26.9% direct comparison weight · independent of the model provider
Epoch AI · 25.9% direct comparison weight · independent of the model provider
ARC Prize · 24.4% direct comparison weight · independent of the model provider
Vals AI · 22.8% direct comparison weight · independent of the model provider

Which benchmark families carry the weight?

ARC-AGI · 24.4% direct comparison weight
GPQA · 7.7% direct comparison weight
Aime · 5.5% direct comparison weight
epoch-game-puzzles · 5.2% direct comparison weight
simpleqa · 4.9% direct comparison weight
Other families (21) · 52.2% 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 2026-10-20.

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	73 → 73	No movement	203 remaining / 241 lost	44.4 (-0.0 points)
ARC Prize	77 → 75	Up 2	444 remaining / 0 lost	49.1 (+4.7 points)
Datacurve	77 → 81	Down 4	444 remaining / 0 lost	39.1 (-5.3 points)
LiveCodeBench	77 → 77	No movement	444 remaining / 0 lost	44.4 (-0.0 points)
Scale	77 → 80	Down 3	444 remaining / 0 lost	46.1 (+1.7 points)
Terminal-Bench	77 → 77	No movement	444 remaining / 0 lost	44.4 (-0.0 points)
Cognition	77 → 79	Down 2	444 remaining / 0 lost	43.3 (-1.1 points)
LiveBench	77 → 79	Down 2	443 remaining / 1 lost	45.1 (+0.7 points)
SWE-rebench	77 → 79	Down 2	444 remaining / 0 lost	44.0 (-0.4 points)
Epoch AI	76 → 84	Down 8	431 remaining / 13 lost	42.0 (-2.4 points)
Vals AI	77 → 71	Up 6	443 remaining / 1 lost	46.9 (+2.5 points)
Provider reports · Anthropic	77 → 77	No movement	444 remaining / 0 lost	44.4 (-0.0 points)
Provider reports · Moonshot AI	77 → 78	Down 1	444 remaining / 0 lost	43.8 (-0.5 points)
Provider reports · Meta	77 → 77	No movement	444 remaining / 0 lost	44.4 (-0.0 points)
Provider reports · Mistral	77 → 77	No movement	444 remaining / 0 lost	44.4 (+0.0 points)
Provider reports · OpenAI	77 → 77	No movement	444 remaining / 0 lost	44.5 (+0.1 points)
All provider reports	77 → 77	No movement	444 remaining / 0 lost	43.9 (-0.5 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
deepswe	39.1	-5.3
arc-agi	48.0	+3.6
epoch-game-puzzles	42.5	-1.9
frontiermath	45.9	+1.5
terminal-bench	45.8	+1.4
mmmu-pro	45.6	+1.2
vibe-code	43.3	-1.1
frontiercode	43.3	-1.1
harvey	43.3	-1.1
mmlu-pro	45.4	+1.0
hle	45.3	+1.0
vals-legal-research	43.6	-0.8
gdp-pdf	45.2	+0.8
vals-excel-modeling	45.1	+0.7
enterprise-ops	43.7	-0.7
vals-code-migration	43.9	-0.5
arc-agi-3	43.9	-0.5
aime	43.9	-0.4
swe-rebench	44.0	-0.4
multi-swe-bench	44.0	-0.4
automationbench	44.1	-0.3
tau-bench	44.1	-0.3
livebench-instructions	44.7	+0.3
simpleqa	44.7	+0.3
vals-finance-agent	44.7	+0.3
swe-atlas-qna	44.1	-0.3
terminal-bench-science	44.7	+0.3
livebench-language	44.6	+0.2
vals-tax-agent	44.6	+0.2
ifbench	44.2	-0.2
livebench-coding	44.2	-0.2
omniscience	44.6	+0.2
swe-atlas-test-writing	44.2	-0.2
gdpval	44.3	-0.1
enigma-eval	44.3	-0.1
livebench-reasoning	44.3	-0.1
apex-agents	44.5	+0.1
livebench-data	44.5	+0.1
itbench	44.3	-0.1
analyst-agent	44.5	+0.1
aa-lcr	44.4	+0.0
swe-atlas-refactoring	44.4	-0.0
livebench-math	44.4	+0.0
gpqa	44.4	+0.0
briefcase	44.4	-0.0
critpt	44.4	-0.0
math500	44.4	+0.0
swe-bench-pro	44.4	+0.0
livecodebench	44.4	+0.0
gmmlu	44.4	+0.0

What evidence is missing from the fit?

Across all collected settings for this model: 90 observations, 45 contributing, 0 matched without graph weight and 45 excluded. Counts do not establish rank eligibility.

3 · Benchmark family or source protocol has not been reviewed
2 · 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
1 · DeepSeek V4 Pro checkpoint is not identified by this source; April and August revisions must not be combined.
2 · DeepSeek V4 Pro checkpoint is unspecified; April and August must remain separate.
2 · Evaluator has not been reviewed for independent admission
14 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately
2 · No matched, identified effort unit; duplicates and unnamed comparators do not add evidence
9 · Supporting evidence outside the reviewed capability core
2 · Narrow specialist biological, medical or security evaluation; preserved as raw evidence outside general capability scoring.
1 · The GLM card attributes this evaluation to Artificial Analysis; it is not a new Z.ai comparison.
1 · Composite index or publisher-specific internal evaluation lacks a stable general-capability comparison unit. Composite index excluded to avoid double counting its ingredients.
6 · Legacy self-report lacks a reviewed comparison configuration

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	Low	High	Max	None	Non-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	Not reported	1,256.94 Elo Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	928.2 Elo Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	80.33% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	50.67% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	56.71% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	34.91% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	18% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	0% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	Not reported
EnterpriseOps-Gym-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Overall task success rate across all eight enterprise domains (oracle tool mode) · Benchmark developed by ServiceNow Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	49.6% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	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	Not reported	11.4% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	3.6% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	1,450.47 Elo Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	1,057.89 Elo Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	92.83% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	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	Not reported	41.01% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	10.57% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	Not reported
MLCR-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of tasks judged accurate, complete, and concise · Expert + compound (hard) sets · Benchmark developed by Wisedocs · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	17.78% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	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	Not reported	0.83 Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	-12.38 Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	49.1% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	41.45% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	51.04% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	39.93% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	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	Not reported	39.59% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	Not reported
Terminal-Bench-Science 0.1Artificial Analysis · mini-swe-agent; Terminal-Bench-Science 0.1.0; 70 tasks; pass@1 averaged over 3 repeats Benchmark details Share of tasks passed (pass@1), averaged over three repeats · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	5.71% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	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	Not reported	78.65% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	Not reported	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	Not reported	14.14% Reported settings DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro	Not reported	1.52% Reported settings DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning	Not reported
ARC-AGI-2ARC Prize · v2 Semi-Private Benchmark details Original public board. Configuration and task coverage may differ; inspect source. Unit: percent. Source collected 2026-10-18.	56.25% Reported settings DeepSeek V4 Pro 0813 (Low); Source model ID: deepseek-v4-pro-0813-low; 120 tasks	59.72% Reported settings DeepSeek V4 Pro 0813 (High); Source model ID: deepseek-v4-pro-0813-high; 120 tasks	61.25% Reported settings DeepSeek V4 Pro 0813 (Max); Source model ID: deepseek-v4-pro-0813-max; 120 tasks	0.83% Reported settings DeepSeek V4 Pro 0813 (None); Source model ID: deepseek-v4-pro-0813-none; 120 tasks	Not reported	Not reported
DeepSWE v1.1Datacurve · mini-swe-agent Benchmark details Original public board. Configuration and task coverage may differ; inspect source. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	62.8% Reported settings deepseek-v4-pro; Reported effort: max; 113 unique tasks; 452 attempts; 4 runs DeepSeek V4 Pro checkpoint is not identified by this source; April and August revisions must not be combined. Source label needs review	Not reported	Not reported	Not reported
FrontierCode 1.1 · extended-chiselCognition · chisel Benchmark details FrontierCode 1.1; chisel. Open each result for its exact effort, task subset, metric and protocol details. Unit: percent. Source collected 2026-10-18.	Not reported	44.45% Reported settings FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.076	Not reported	Not reported	Not reported	Not reported
FrontierCode 1.1 · extended-mini-swe-agentCognition · mini-swe-agent Benchmark details FrontierCode 1.1; mini-swe-agent. Open each result for its exact effort, task subset, metric and protocol details. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	Not reported	Not reported	Not reported	31.01% Reported settings FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067 DeepSeek V4 Pro checkpoint is unspecified; April and August must remain separate. Source label needs review
FrontierCode 1.1 · main-chiselCognition · chisel Benchmark details FrontierCode 1.1; chisel. Open each result for its exact effort, task subset, metric and protocol details. Unit: percent. Source collected 2026-10-18.	Not reported	28.55% Reported settings FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.106	Not reported	Not reported	Not reported	Not reported
FrontierCode 1.1 · main-mini-swe-agentCognition · mini-swe-agent Benchmark details FrontierCode 1.1; mini-swe-agent. Open each result for its exact effort, task subset, metric and protocol details. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	Not reported	Not reported	Not reported	17.64% Reported settings FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.01 DeepSeek V4 Pro checkpoint is unspecified; April and August must remain separate. Source label needs review
Epoch · Chess Puzzles · v2.0.0Epoch AI · Epoch AI Inspect; task 2.0.0; scorer exact Benchmark details Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	46% Reported settings Epoch AI Inspect; task 2.0.0; scorer exact; model=deepseek-v4-pro-0813_max; run=C56VHYLMeww2wXebeJh3gg; mean accuracy; https://epoch.ai/data/benchmarks.csv	Not reported	Not reported	Not reported
Epoch · FrontierMath Tier 4 · v2.0.0Epoch AI · Epoch AI Inspect; task 2.0.0; scorer verification_code Benchmark details Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY. FrontierMath was developed with OpenAI funding and unequal access to part of the question set; this result is from Epoch's evaluation. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	26.83% Reported settings Epoch AI Inspect; task 2.0.0; scorer verification_code; model=deepseek-v4-pro-0813_max; run=a4uV8HMMBAcHXLVSmZfBgX; mean accuracy; https://epoch.ai/data/benchmarks.csv	Not reported	Not reported	Not reported

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 · 8,915 results across 259 models · Collected 2026-10-18

DeepSeek V4-Pro 0813 · Max

Compare with

One configuration, seven capabilities

Agentic

Hard reasoning

Coding

Human pref

Knowledge

Multimodal

Long context

Capability score evidence

Benchmark results