RankingEffort benchmarks
Gemini 2.5 Flash Preview
15 measured results · 6 effort settings · 7 benchmark/harness combinations
Change model
Capability profile
One configuration, seven capabilities
- AGemini 2.5 Flash Preview · Reasoning1 of 7 capabilities supported
Gaps are unknown, not zero. Hollow points are preliminary.
Agentic
UnknownHard reasoning
A9.6SupportedCoding
A3.8PreliminaryHuman pref
UnknownKnowledge
A7.2PreliminaryMultimodal
UnknownLong context
Unknown
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
Gemini 2.5 Flash Preview · Reasoning: 2.2 overall comparison score. Meets the overall evidence requirements.
Shared-result disagreement. This aggregate order conflicts with matched benchmark results against 1 other configurations. Different test coverage and opponents affect the fit. Inspect published results before treating the rank as conclusive.
Which sources carry the weight?
- Artificial Analysis · 100.0% direct comparison weight · independent of the model provider
Which benchmark families carry the weight?
- MMLU-Pro · 20.6% direct comparison weight
- math500 · 20.4% direct comparison weight
- LiveCodeBench · 19.9% direct comparison weight
- GPQA · 19.8% direct comparison weight
- Humanity’s Last Exam · 19.2% 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 2026-11-22.
Both ranks use the same surviving configurations for each removal. The cohort starts with 602 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.
| Removed source | Rank before → after | Rank movement | Cohort remaining / lost support | Score after (change) |
|---|---|---|---|---|
| Artificial Analysis | Comparison support lost; no rank | 230 remaining / 372 lost | Unavailable | |
| ARC Prize | 387 → 385 | Up 2 | 602 remaining / 0 lost | 2.9 |
| Datacurve | 387 → 387 | No movement | 602 remaining / 0 lost | 1.8 |
| LiveCodeBench | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
| Scale | 387 → 387 | No movement | 602 remaining / 0 lost | 2.5 |
| Terminal-Bench | 387 → 388 | Down 1 | 602 remaining / 0 lost | 2.2 |
| Cognition | 387 → 388 | Down 1 | 602 remaining / 0 lost | 2.1 |
| LiveBench | 386 → 386 | No movement | 601 remaining / 1 lost | 2.4 |
| SWE-rebench | 387 → 388 | Down 1 | 602 remaining / 0 lost | 2.3 |
| Epoch AI | 376 → 376 | No movement | 588 remaining / 14 lost | 2.3 |
| Vals AI | 386 → 386 | No movement | 601 remaining / 1 lost | 2.2 |
| Provider reports · Anthropic | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
| Provider reports · Moonshot AI | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
| Provider reports · Meta | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
| Provider reports · Mistral | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
| Provider reports · OpenAI | 387 → 387 | No movement | 602 remaining / 0 lost | 2.3 |
| All provider reports | 387 → 387 | No movement | 602 remaining / 0 lost | 2.2 |
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 | 2.8 | +0.6 |
| livecodebench | 2.8 | +0.5 |
| deepswe | 1.8 | -0.5 |
| math500 | 1.9 | -0.3 |
| hle | 2.0 | -0.3 |
| mmlu-pro | 2.5 | +0.3 |
| mmmu-pro | 2.4 | +0.2 |
| gpqa | 2.4 | +0.2 |
| critpt | 2.1 | -0.1 |
| epoch-game-puzzles | 2.1 | -0.1 |
| simpleqa | 2.4 | +0.1 |
| frontiercode | 2.1 | -0.1 |
| aa-lcr | 2.1 | -0.1 |
| tau-bench | 2.2 | -0.1 |
| ifbench | 2.2 | -0.1 |
| livebench-instructions | 2.3 | +0.1 |
| swe-rebench | 2.3 | +0.1 |
| enigma-eval | 2.3 | +0.1 |
| multi-swe-bench | 2.2 | -0.0 |
| arc-agi-3 | 2.2 | -0.0 |
| enterprise-ops | 2.2 | -0.0 |
| gdp-pdf | 2.3 | +0.0 |
| apex-agents | 2.3 | +0.0 |
| livebench-language | 2.3 | +0.0 |
| terminal-bench | 2.3 | +0.0 |
| gdpval | 2.3 | +0.0 |
| livebench-coding | 2.2 | -0.0 |
| terminal-bench-science | 2.3 | +0.0 |
| swe-atlas-qna | 2.2 | -0.0 |
| aime | 2.2 | -0.0 |
| vibe-code | 2.2 | -0.0 |
| analyst-agent | 2.3 | +0.0 |
| livebench-reasoning | 2.2 | -0.0 |
| vals-code-migration | 2.2 | -0.0 |
| itbench | 2.2 | -0.0 |
| vals-legal-research | 2.2 | -0.0 |
| vals-finance-agent | 2.2 | -0.0 |
| briefcase | 2.2 | -0.0 |
| automationbench | 2.2 | -0.0 |
| vals-tax-agent | 2.2 | -0.0 |
| swe-bench-pro | 2.3 | +0.0 |
| swe-atlas-test-writing | 2.2 | -0.0 |
| vals-excel-modeling | 2.2 | -0.0 |
| frontiermath | 2.2 | -0.0 |
| harvey | 2.2 | +0.0 |
| livebench-data | 2.2 | +0.0 |
| omniscience | 2.2 | +0.0 |
| swe-atlas-refactoring | 2.2 | -0.0 |
| gmmlu | 2.2 | -0.0 |
| livebench-math | 2.2 | +0.0 |
What evidence is missing from the fit?
Across all collected settings for this model: 15 observations, 15 contributing, 0 matched without graph weight and 0 excluded. Counts do not establish rank eligibility.
Benchmark results
Reviewed retained observations are in use. October 2026 Real-SWE board has no 95% confidence whisker, so these scores stay at the 2026-10-18 retrieval. Original source retrieval 2026-10-18; source updated 2026-10-18.
| Benchmark / source | Thinking 1K | Thinking 8K | Thinking 16K | Thinking 24K | Reasoning | Non-reasoning |
|---|---|---|---|---|---|---|
GPQA DiamondBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | 69.8% Reported settingsGemini 2.5 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning-04-2025 | 59.39% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
Humanity's Last ExamBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | 12.1% Reported settingsGemini 2.5 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning-04-2025 | 3.8% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
LiveCodeBenchBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | 50.48% Reported settingsGemini 2.5 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning-04-2025 | 40.63% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
MATH-500Benchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | 98.07% Reported settingsGemini 2.5 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning-04-2025 | 92.6% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
MMLU-ProBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | 79.98% Reported settingsGemini 2.5 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning-04-2025 | 78.31% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
MMMU-ProBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | Not reported | Not reported | Not reported | Not reported | Not reported | 61.97% Reported settingsGemini 2.5 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-04-2025 |
ARC-AGI-2Benchmark detailsOriginal public board. Configuration and task coverage may differ; inspect source. Unit: percent. Source collected 2026-11-22. | 2.16% Reported settingsGemini 2.5 Flash (Preview) (Thinking 1K); Source model ID: gemini-2-5-flash-preview-05-20-thinking-1k; 120 tasks | 2.12% Reported settingsGemini 2.5 Flash (Preview) (Thinking 8K); Source model ID: gemini-2-5-flash-preview-05-20-thinking-8k; 120 tasks | 1.98% Reported settingsGemini 2.5 Flash (Preview) (Thinking 16K); Source model ID: gemini-2-5-flash-preview-05-20-thinking-16k; 120 tasks | 2.54% Reported settingsGemini 2.5 Flash (Preview) (Thinking 24K); Source model ID: gemini-2-5-flash-preview-05-20-thinking-24k; 120 tasks | 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 · 10,184 results across 381 models · Collected 2026-11-22