DeepSeek: DeepSeek V4 Flash Vision Exp
DeepSeek V4 Flash Vision Exp is an experimental vision-enabled version of [DeepSeek V4 Flash 0731](https://openrouter.ai/deepseek/deepseek-v4-flash-0731) from DeepSeek, adding image understanding whi…
deepseek/deepseek-v4-flash-vision-exp-20260821
DeepSeek V4 Flash Vision Exp is an experimental vision-enabled version of [DeepSeek V4 Flash 0731](https://openrouter.ai/deepseek/deepseek-v4-flash-0731) from DeepSeek, adding image understanding while matching the base model on text capabilities including agents, reasoning, and world knowledge. It is a sparse mixture-of-experts model with 13B active parameters out of 284B total.
It is suited for document and chart understanding, visual question answering, and multimodal agent workflows that interleave text and images.
Specifications & capabilities
- Input
- text, image
- Output
- text
- Context window · tokens
- 1048576
- Published model ID
- DeepSeek-V4-Flash-Vision-Exp
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Description source ↗Evaluations
1–17 / 17
90.909%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · GMICloud
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 2
- sample size
- 1386
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 7
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- GMICloud
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-11-10T06:07:30.895Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 1386
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.909090857142857
- accuracy_std_dev
- 0.0075824593339510055
- cost_per_task_usd
- 0.038180986291486294
- output_tokens_per_task
- 28692.146464646463
- latency_seconds_per_task
- 147.7676544011544
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 26
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗89.815%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · SiliconFlow
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 5
- sample size
- 1188
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 6
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- SiliconFlow
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-03T09:28:24.400Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 1188
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.898148
- accuracy_std_dev
- 0.012012972737836369
- cost_per_task_usd
- 0.03667170117845118
- output_tokens_per_task
- 27519.53787878788
- latency_seconds_per_task
- 157.79789814814814
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 34
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗76%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · SiliconFlow
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 5
- sample size
- 200
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 4
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- SiliconFlow
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-03T08:13:12.597Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 200
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.76
- accuracy_std_dev
- 0.01333300000000004
- cost_per_task_usd
- 0.054461335
- output_tokens_per_task
- 15101.33
- latency_seconds_per_task
- 91.417
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 28
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗89.944%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · Novita
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 7
- sample size
- 1185
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 6
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- Novita
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-03T09:28:39.969Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 1185
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.8994372877637131
- accuracy_std_dev
- 0.01977995405637288
- cost_per_task_usd
- 0.03544181097046414
- output_tokens_per_task
- 26597.288607594935
- latency_seconds_per_task
- 152.64521687763715
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 59
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗76.667%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · Novita
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 7
- sample size
- 50
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 1
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- Novita
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-09-01T12:41:31.535Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 50
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.7666669999999999
- cost_per_task_usd
- 0.0337152
- output_tokens_per_task
- 17702.16
- latency_seconds_per_task
- 192.2459
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 52
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗86.322%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · Fireworks
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 9
- sample size
- 792
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 4
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- Fireworks
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-09-19T08:51:58.825Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 792
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.86321525
- accuracy_std_dev
- 0.010127171825712258
- cost_per_task_usd
- 0.022006194444444443
- output_tokens_per_task
- 33108.22601010101
- latency_seconds_per_task
- 265.79813636363633
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 17
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗76.222%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · DeepInfra
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 12
- sample size
- 300
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 6
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- DeepInfra
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-08T22:27:05.549Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 300
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.7622223333333333
- accuracy_std_dev
- 0.030469494967188978
- cost_per_task_usd
- 0.029655940000000006
- output_tokens_per_task
- 18856.086666666666
- latency_seconds_per_task
- 256.38588
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 63
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗74.095%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · GMICloud
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 12
- sample size
- 350
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 7
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- GMICloud
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-11-10T06:15:52.514Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 350
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.7409525714285714
- accuracy_std_dev
- 0.01094194948492327
- cost_per_task_usd
- 0.03426931142857143
- output_tokens_per_task
- 15886.077142857142
- latency_seconds_per_task
- 134.51465142857143
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 23
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗88.119%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · DeepInfra
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 13
- sample size
- 1386
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 7
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- DeepInfra
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-08T22:18:51.642Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 1386
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.8811927142857143
- accuracy_std_dev
- 0.007888829783113527
- cost_per_task_usd
- 0.015891973304473304
- output_tokens_per_task
- 24276.554834054834
- latency_seconds_per_task
- 279.40710173160176
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 71
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗67.778%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · Fireworks
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 14
- sample size
- 150
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 3
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- Fireworks
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-09-19T08:01:53.393Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 150
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.6777776666666667
- accuracy_std_dev
- 0.013698571612479227
- cost_per_task_usd
- 0.0340762
- output_tokens_per_task
- 27086.806666666667
- latency_seconds_per_task
- 244.09344000000002
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 14
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗73.333%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · AtlasCloud
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 15
- sample size
- 50
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 1
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- AtlasCloud
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-09-12T08:09:12.993Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 50
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.7333330000000001
- cost_per_task_usd
- 0.058005
- output_tokens_per_task
- 30812.32
- latency_seconds_per_task
- 191.24362
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 29
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗83.389%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · AtlasCloud
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 18
- sample size
- 594
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 3
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- AtlasCloud
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-09-12T09:56:44.384Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 594
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.8338943333333333
- accuracy_std_dev
- 0.011690806112307073
- cost_per_task_usd
- 0.03773407239057239
- output_tokens_per_task
- 28290.345117845118
- latency_seconds_per_task
- 276.3961228956229
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 30
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗74.8%
- source id
- openRouterBenchmarks
- leaderboard name
- tau2-bench Verified Airline · Accuracy · auto-routing
- dimension
- agentic
- metric
- tau2-bench Verified Airline · Accuracy
- rank
- 34
- sample size
- 250
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 5
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- tau_bench_verified_airline
- name
- tau2-bench Verified Airline
- record_kind
- aggregate
- provider_name
- auto-routing
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-03T08:18:33.163Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- tau2-bench Verified Airline · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 250
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/tau2-bench-airline
- measurements
- accuracy
- 0.7480002
- accuracy_std_dev
- 0.020396130352593862
- cost_per_task_usd
- 0.030171100000000003
- output_tokens_per_task
- 18663.832
- latency_seconds_per_task
- 197.857312
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 128
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗88.019%
- source id
- openRouterBenchmarks
- leaderboard name
- GPQA Diamond · Accuracy · auto-routing
- dimension
- knowledge
- metric
- GPQA Diamond · Accuracy
- rank
- 35
- sample size
- 1188
- aggregation scope
- status
- published
- harness
- url
- https://github.com/OpenRouterTeam/benchmark-harness
- name
- OpenRouter benchmark harness
- direction
- higherIsBetter
- run_count
- 6
- conditions
- window
- newest 90 days; successful visible nonzero runs
- dataset
- benchmark-leaderboard
- benchmark
- id
- gpqa_diamond
- name
- GPQA Diamond
- record_kind
- aggregate
- provider_name
- auto-routing
- attribution
- Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
- last_run_at
- 2026-10-03T09:30:18.263Z
- license_url
- https://creativecommons.org/licenses/by/4.0/
- methodology
- Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
- metric_kind
- GPQA Diamond · Accuracy
- rank_method
- competition rank among reported evaluations in identical conditions, including unlinked observations
- sample_size
- 1188
- source_file
- benchmark-leaderboard.json
- source_page
- https://openrouter.ai/benchmarks/gpqa-diamond
- measurements
- accuracy
- 0.8801906666666667
- accuracy_std_dev
- 0.011226983784119793
- cost_per_task_usd
- 0.0227881835016835
- output_tokens_per_task
- 25264.190235690236
- latency_seconds_per_task
- 234.70068434343435
- published_at
- 2026-11-16T01:46:02.653Z
- snapshot_date
- 2026-11-16
- source_revision
- 7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
- transformations
- TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
- metric_population
- 132
- benchmark_evaluation
- true
- openrouter_permaslug
- deepseek/deepseek-v4-flash-vision-exp-20260821
- source model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
Original result ↗65
- source id
- artificialAnalysis
- dimension
- coding
- metric
- Coding Index
Original result ↗47.5
- source id
- artificialAnalysis
- dimension
- agentic
- metric
- Agentic Index
Original result ↗34.8
- source id
- artificialAnalysis
- dimension
- intelligence
- metric
- Intelligence Index
Original result ↗Published prices
1–15 / 15
$0.01
- Inference provider
- openRouter
- Platform
- OpenRouter
- Unit
- cache_read_tokens_per_million
- Currency
- USD
$0.44
- Inference provider
- openRouter
- Platform
- OpenRouter
- Unit
- input_tokens_per_million
- Currency
- USD
$1.32
- Inference provider
- openRouter
- Platform
- OpenRouter
- Unit
- output_tokens_per_million
- Currency
- USD
$0.03
- Inference provider
- siliconFlow
- Platform
- SiliconFlow
- Unit
- cache_read_tokens_per_million
- Currency
- USD
- Billing label
- Cached input tokens (per 1M)
$0.44
- Inference provider
- siliconFlow
- Platform
- SiliconFlow
- Unit
- input_tokens_per_million
- Currency
- USD
- Billing label
- Input tokens (per 1M)
$1.32
- Inference provider
- siliconFlow
- Platform
- SiliconFlow
- Unit
- output_tokens_per_million
- Currency
- USD
- Billing label
- Output tokens (per 1M)
$0.01
- Inference provider
- vercelAIGatewayCatalog
- Platform
- deepinfra
- Unit
- cache_read_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- deepinfra | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- quantization
- fp8
- billing_scope
- via Vercel AI Gateway
- Billing label
- Cached input / 1M
- Pricing via
- Vercel AI Gateway
$0.22
- Inference provider
- vercelAIGatewayCatalog
- Platform
- deepinfra
- Unit
- input_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- deepinfra | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- quantization
- fp8
- billing_scope
- via Vercel AI Gateway
- Billing label
- Input tokens / 1M
- Pricing via
- Vercel AI Gateway
$0.65
- Inference provider
- vercelAIGatewayCatalog
- Platform
- deepinfra
- Unit
- output_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- deepinfra | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- quantization
- fp8
- billing_scope
- via Vercel AI Gateway
- Billing label
- Output tokens / 1M
- Pricing via
- Vercel AI Gateway
$0.03
- Inference provider
- vercelAIGatewayCatalog
- Platform
- novita
- Unit
- cache_read_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- novita | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- billing_scope
- via Vercel AI Gateway
- Billing label
- Cached input / 1M
- Pricing via
- Vercel AI Gateway
$0.44
- Inference provider
- vercelAIGatewayCatalog
- Platform
- novita
- Unit
- input_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- novita | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- billing_scope
- via Vercel AI Gateway
- Billing label
- Input tokens / 1M
- Pricing via
- Vercel AI Gateway
$1.32
- Inference provider
- vercelAIGatewayCatalog
- Platform
- novita
- Unit
- output_tokens_per_million
- Currency
- USD
- Billing conditions
- endpoint
- novita | deepseek/deepseek-v4-flash-vision-exp
- model_type
- language
- billing_scope
- via Vercel AI Gateway
- Billing label
- Output tokens / 1M
- Pricing via
- Vercel AI Gateway
$0.01
- Inference provider
- vercelAIGatewayCatalog
- Platform
- Vercel AI Gateway
- Unit
- cache_read_tokens_per_million
- Currency
- USD
- Billing conditions
- model_type
- language
- quote_scope
- gateway_catalog
- billing_scope
- Vercel public catalog; endpoint prices may differ
- varies_by_provider
- true
- Billing label
- Cached input / 1M
- Pricing via
- Vercel AI Gateway
$0.22
- Inference provider
- vercelAIGatewayCatalog
- Platform
- Vercel AI Gateway
- Unit
- input_tokens_per_million
- Currency
- USD
- Billing conditions
- model_type
- language
- quote_scope
- gateway_catalog
- billing_scope
- Vercel public catalog; endpoint prices may differ
- varies_by_provider
- true
- Billing label
- Input tokens / 1M
- Pricing via
- Vercel AI Gateway
$0.65
- Inference provider
- vercelAIGatewayCatalog
- Platform
- Vercel AI Gateway
- Unit
- output_tokens_per_million
- Currency
- USD
- Billing conditions
- model_type
- language
- quote_scope
- gateway_catalog
- billing_scope
- Vercel public catalog; endpoint prices may differ
- varies_by_provider
- true
- Billing label
- Output tokens / 1M
- Pricing via
- Vercel AI Gateway
Provider specifications & access
1–3 / 3
- source model id
- deepseek-ai/DeepSeek-V4-Flash-Vision-Exp
- hosting
- siliconflow-hosted
- max output tokens
- 393216
- capabilities
- promptCaching
- reasoning
- structuredOutputs
- toolCalling
- source model id
- deepseek/deepseek-v4-flash-vision-exp
- hosting
- vercel-gateway
- context length
- 1048576
- max output tokens
- 1048576
- capabilities
- promptCaching
- reasoning
- structuredOutputs
- toolCalling
- visionInput
- source model id
- publishers/deepseek-ai/models/deepseek-v4@deepseek-v4-flash-vision-exp
- hosting
- vertex-self-deployed
Safety assessments
1–1 / 1
32 / 100
- source id
- stanfordFMTI
- canonical model id
- deepseek/deepseek-v4-flash-vision-exp-20260821
- taxonomy
- transparency
- hazard category
- transparency
- hazard name
- Organization transparency
- policy area
- DeepSeek
- assessment type
- organization-level disclosure
- metric
- disclosure_score
- result label
- 32 / 100
- sample size
- 100
- updated at
- 2026-11-09T18:08:05.227Z
Source specification observations
1–6 / 6
- field
- context_length
- value
- 1048576
- normalized value
- 1048576
- scope
- source-model
- source id
- openRouter
- confidence
- 1
- field
- context_length
- value
- 1049K
- scope
- source-model
- source id
- siliconFlow
- confidence
- 1
- field
- context_length
- value
- 1048576
- normalized value
- 1048576
- scope
- source-model
- source id
- vercelAIGatewayCatalog
- confidence
- 1
- field
- max_output_tokens
- value
- 943717
- normalized value
- 943717
- scope
- source-model
- source id
- openRouter
- confidence
- 1
- field
- max_output_tokens
- value
- 393K
- scope
- source-model
- source id
- siliconFlow
- confidence
- 1
- field
- max_output_tokens
- value
- 1048576
- normalized value
- 1048576
- scope
- source-model
- source id
- vercelAIGatewayCatalog
- confidence
- 1
Hugging Face model card
1–1 / 1
- downloads
- 835578
- likes
- 932
- pipeline tag
- image-text-to-text
- library name
- transformers
- License
- mit
Explore more models