DeepSeek v4 Flash (max): benchmark scores, price and ranking

DeepSeek v4 Flash (max) from DeepSeek ranks #255 of 743 LLMs on the LLMs Tiger aggregate benchmark score (49.6), based on 5 benchmark results captured up to 2026-10-13.

Organisation: DeepSeek
Type: Open-weight / local model
Open weights: Yes
Released: 2026-04-24 (source: epoch)
Cheapest API price: $0.090 in / $0.180 out per million tokens (via deepinfra, observed 2026-10-13T20:47Z)
Aggregate rank: #255 of 743 (score 49.6)

Relative to other models, DeepSeek v4 Flash (max) is strongest on DTBench (#36 of 146) and weakest on SciCode (#52 of 106).

Across the 5 benchmarks where at least ten models are scored, DeepSeek v4 Flash (max) is in the top 10% on 0 and in the top half on 5.

Its cheapest listed input price, $0.090 per million tokens, is lower than that of 83% of the 427 models with a listed price.

DeepSeek v4 Flash (max) benchmark results

DeepSeek v4 Flash (max) benchmark scores, ranked against all models measured on each benchmark
Benchmark	Score	Rank	Source	Captured
DTBench	86.4%	#36 of 146	Epoch AI Benchmarking Hub	2026-10-13
WeirdML	45.6%	#39 of 109	Epoch AI Benchmarking Hub	2026-10-13
CritPt	7.1%	#45 of 115	Epoch AI Benchmarking Hub	2026-10-13
LMCA	35.9%	#43 of 108	Epoch AI Benchmarking Hub	2026-10-13
SciCode	44.9%	#52 of 106	Epoch AI Benchmarking Hub	2026-10-13

About DeepSeek v4 Flash (max)

DeepSeek v4 Flash (max) is an open-weight model that can run on your own hardware from DeepSeek. It was released on 2026-04-24, according to Epoch AI. The cheapest API listing we track charges $0.090 per million input tokens and $0.180 per million output tokens, through deepinfra. We hold 5 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where DeepSeek v4 Flash (max) stands

On our aggregate score, DeepSeek v4 Flash (max) is #255 of 743 ranked models, around the middle of the table. Its score is 49.6; the scale is centred on 50, so it sits 0.4 points below the typical model. That beats 66% of the ranked field. Among the 33 ranked DeepSeek models it is #13. The next one up is DeepSeek Coder V2 0724 at 49.9. The next one down is deepseek-v4-pro-high-20260813 at 48.9.

DeepSeek v4 Flash (max), benchmark by benchmark

DTBench puts DeepSeek v4 Flash (max) at 86.4%, #36 of 146. That beats 75% of the models tested, but 12.5 points short of Claude Opus 5.5 (max) (98.9%). The results either side are Qwen 3.6 Max (Preview) (87.2%) and DeepSeek-V3.2 (Thinking; Novita) (85.6%). It covers dynamic tool-use benchmark. The board median is 68.8%, so DeepSeek v4 Flash (max) is 17.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On WeirdML, DeepSeek v4 Flash (max) reaches 45.6%. It ranks #39 of 109, ahead of 64% of the field, and 47.6 points behind GPT-6 Astra (max) at 93.3%. The results either side are grok-4-0709 (45.7%) and Kimi K2.5 (45.6%). It covers unusual machine-learning tasks. The board median is 41.2%, so DeepSeek v4 Flash (max) is 4.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

DeepSeek v4 Flash (max) manages 7.1% on CritPt, good for #45 of 115. GPT-5.6 Sol (max) leads at 32.3%, 25.1 points further up, and 61% of models sit below. The results either side are Kimi K2.6 (8.0%) and Gemini 3 Pro Preview (6.9%). It covers unpublished research-level physics problems. The board median is 2.9%, so DeepSeek v4 Flash (max) is 4.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

LMCA puts DeepSeek v4 Flash (max) at 35.9%, #43 of 108. That beats 60% of the models tested, but 32.4 points short of Claude Opus 5.5 (max) (68.2%). The results either side are Qwen3.5 Plus (36.4%) and Qwen3.6 27B (34.5%). It covers language-model capability assessment. The board median is 29.2%, so DeepSeek v4 Flash (max) is 6.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On SciCode, DeepSeek v4 Flash (max) reaches 44.9%. It ranks #52 of 106, ahead of 51% of the field, and 22.0 points behind Claude Opus 5.5 (max) at 66.9%. The results either side are GLM-4.7 (45.1%) and Muse Glimmer (high) (44.9%). It covers scientific research code. The board median is 44.3%, so DeepSeek v4 Flash (max) is 0.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Strengths and weaknesses

General knowledge and chat: solid, ahead of 62% of the field on average across 2 benchmarks. Its best board here is WeirdML (#39 of 109) and its weakest is LMCA (#43 of 108).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so DeepSeek v4 Flash (max) costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), DeepSeek V4 Flash 0731 (max) ($0.016, 53.1). Similar scores cost more on Reka Flash 3 ($0.100) and Qwen3.8 Flash (xhigh) ($0.113).

Release history

The DeepSeek model released before it was DeepSeek-V3.2-Speciale on 2025-12-01, which scores 50.4. The next one was DeepSeek v4 Pro (max) on 2026-04-24, at 51.6.

Reading these numbers

The 5 scores on this page come from 1 source: Epoch AI Benchmarking Hub (5). No official leaderboard supplied any of them, so treat them as third-party measurements. The leaderboard also lists 2 other reasoning-effort settings of this model, which redirect here. This page shows the highest. With 5 results the aggregate rank can move a lot when one new score arrives.

DeepSeek v4 Flash (max) is also listed at 2 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

#252 qwen3.5-4b (none) (49.6)
#253 Qwen3.5 Plus (49.6)
#254 Mistral-Large-Instruct-2411 (49.6)
#256 gpt-4o-2024-11-20 (49.6)
#257 claude-haiku-4-5-20251001_16K (49.5)
#258 Kimi-K2-Instruct-0905 (49.5)