vs
1:6
overall capability wins (0 ties across 7 scored). DeepSeek V4.1 Flash leads overall.
- Reasoning0:3
- Coding0:1
- Composite indices0:2
- Pricing3:0
- Specs1:2
Verdict
DeepSeek V4.1 Flash leads coding (widest gap: +16.1 on SciCode); DeepSeek V4.1 Flash offers ~7.8x larger context (1M vs 128K); DeepSeek V4.1 Flash allows ~46.9x more max output (384K vs 8K).
- DeepSeek V4.1 Flash leads coding (widest gap: +16.1 on SciCode)
- DeepSeek V4.1 Flash offers ~7.8x larger context (1M vs 128K)
- DeepSeek V4.1 Flash allows ~46.9x more max output (384K vs 8K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- DeepSeek V3
- $0.55
- DeepSeek V4.1 Flash
- $0.60
DeepSeek V3 is about 1.1x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
DeepSeek
DeepSeek
Family
DeepSeek
DeepSeek
License
MIT
MIT
Open weights
Yes
Yes
Release
Dec 26, 2024
Sep 10, 2026
Knowledge cutoff
-
-
API / provider
DeepSeek
DeepSeek
Modalities
text → text
text, image → text
Specs
Context window
128K
1M
Max output
8K
384K
Parameters
671B (37B act.)
552B (16B act.)
Pricing
Input $/1M
$0.27
$0.30
Output $/1M
$1.10
$1.20
Blended $/1M (3∶1)
$0.48
$0.52
Speed
tok/s
80
-
TTFT (s)
0.4
-
Reasoning
MMLU-Pro
75.9
—
GPQA Diamond
55.7
—
Humanity's Last Exam
2.9
39.2
AIME 2025
26.0
—
MATH-500
90.2
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
25.5
46.4
AA-LCR v1.1
29.3
84.0
CritPt
—
14.3
MMMU-Pro
—
77.0
IFBench
34.8
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
49.6
—
Terminal-Bench 2.1
16.9
—
Aider Polyglot
55.1
—
Terminal-Bench 3
—
—
BigCodeBench
50.0
—
SciCode
35.8
51.9
CursorBench
—
—
SWE-Rebench
23.3
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,620
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,358
—
| Metric | DeepSeek V3 | DeepSeek V4.1 Flash | Delta |
|---|---|---|---|
| Identity | |||
| Organization | DeepSeek | DeepSeek | — |
| Family | DeepSeek | DeepSeek | — |
| License | MIT | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Dec 26, 2024 | Sep 10, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | DeepSeek | DeepSeek | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 1M | B +872K |
| Max output | 8K | 384K | B +376K |
| Parameters | 671B (37B act.) | 552B (16B act.) | A +119 |
| Pricing | |||
| Input $/1M | $0.27 | $0.30 | A +$0.03 |
| Output $/1M | $1.10 | $1.20 | A +$0.10 |
| Blended $/1M (3∶1) | $0.48 | $0.52 | A +$0.05 |
| Speed | |||
| tok/s | 80 | - | — |
| TTFT (s) | 0.4 | - | — |
| Reasoning | |||
| MMLU-Pro | 75.9 | — | — |
| GPQA Diamond | 55.7 | — | — |
| Humanity's Last Exam | 2.9 | 39.2 | B +36.3 pts |
| AIME 2025 | 26.0 | — | — |
| MATH-500 | 90.2 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 25.5 | 46.4 | B +20.9 pts |
| AA-LCR v1.1 | 29.3 | 84.0 | B +54.7 pts |
| CritPt | — | 14.3 | — |
| MMMU-Pro | — | 77.0 | — |
| IFBench | 34.8 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 49.6 | — | — |
| Terminal-Bench 2.1 | 16.9 | — | — |
| Aider Polyglot | 55.1 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | 50.0 | — | — |
| SciCode | 35.8 | 51.9 | B +16.1 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 23.3 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,620 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,358 | — | — |
DeepSeek V3
- Context
- 128K / 8K out
- Parameters
- 671B (37B act.)
- Price
- $0.27 / $1.10
- Speed
- 80 tok/s · 0.4s TTFT
- Modalities
- text → text
- License
- MIT
Original DeepSeek V3 MoE - the release that reset open-weight price/performance expectations.
DeepSeek V4.1 Flash
- Context
- 1M / 384K out
- Parameters
- 552B (16B act.)
- Price
- $0.30 / $1.20 peak · $0.15 / $0.60 off-peak
- Speed
- —
- Modalities
- text, image → text
- License
- MIT
New default DeepSeek Flash tier (Sep 10, 2026), the smallest model in a new Causal Encoder-Decoder architecture family: 552B total parameters activating 8B (prefill) / 16B (decode) per token, MIT-licensed weights on Hugging Face. Replaces DeepSeek-V4-Flash and V4-Flash-Vision-Exp, which are retired but still routed to this model for compatibility. DeepSeek's own reported numbers (GPQA Diamond 90.9, Terminal-Bench 2.1 90.6, text-only HLE 39.1) are self-measured and not directly comparable to this catalog's Artificial-Analysis-sourced cells. API pricing: $0.30 in / $1.20 out peak, $0.15 in / $0.60 out off-peak (per 1M tokens); off-peak cache-hit input is $0.003.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.