DeepSeek V3
MIT Open SourceDeveloped by DeepSeek · MoE (Multi-head Latent Attention) · Modalities: Text · Released 2024-12-26
What are the token costs and operational benchmarks for DeepSeek V3?
Verified Capability Benchmarks
Composite Intelligence: 63.5/100Architectural & Production Profile
671B parameter Mixture-of-Experts open-weight model (37B active) delivering strong price-to-performance for general text workloads.
Cost-sensitive batch text processing, drafting, and open-weight self-hosted inference.
- Architecture
- MoE (Multi-head Latent Attention)
- Modalities
- Text
- Max Output Tokens
- 8,192 tokens
- Blended Rate (3:1)
- $0.175 / 1M
Verified Cloud Inference Endpoints Hosting DeepSeek V3
Compare latency, streaming speed, and token pricing across cloud API providers.
| Inference Provider | Role | Input / 1M | Output / 1M | Avg TTFT | Throughput |
|---|---|---|---|---|---|
| DeepSeek Platform → | Primary Host | $0.14 | $0.28 | 410ms | 78 tok/s |
| Alibaba Cloud DashScope → | Partner Endpoint | $0.14 | $0.28 | 370ms | 85.8 tok/s |
Interactive Monthly Token Economics & ROI Forecaster
Model production workloads across prompt, completion, and cached context tokens, or switch models on the fly.
Estimated Monthly Spend
$0.14/1M in · $0.28/1M out
Prompt Cache: $0.014/1M read · $0.14/1M write
$2.00/1M in · $10.00/1M out
Prompt Cache: $0.100/1M read · $0.50/1M write
Head-to-Head Comparisons Involving DeepSeek V3 (36)
DeepSeek V3 vs Qwen 3.8 Flash Next
Qwen 3.8 Flash Next delivers superior capability and speed, outclassing DeepSeek V3 despite higher pricing. Qwen dominates SWE-bench Verified (81.0% vs. 42.0%) and GPQA Diamond (91.7% vs. 59.1%), alongside a 280ms TTFT (130ms faster) and 185 vs. 78 tok/s throughput. DeepSeek V3 is 12.5% cheaper on input ($0.14 vs. $0.16) and $0.28 vs. $0.47 for output.
DeepSeek V3 vs Gemini 4 Argon
DeepSeek V3 is 93.0% cheaper for input tokens ($0.14 vs. $2.00 per 1M tokens) and $0.28 vs. $10.00 for output tokens (14.3x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (440ms faster than Gemini 4 Argon) with 78 vs. 115 tok/s throughput. Gemini 4 Argon leads coding evaluations at 84.5% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 93.8% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Claude 4.5 Sonnet
DeepSeek V3 is 95.3% cheaper for input tokens ($0.14 vs. $3.00 per 1M tokens) and $0.28 vs. $15.00 for output tokens (21.4x cost advantage). In terms of operational performance, Claude 4.5 Sonnet delivers lower response latency at 380ms TTFT (30ms faster than DeepSeek V3) with 84 vs. 78 tok/s throughput. Claude 4.5 Sonnet leads coding evaluations at 75.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 84.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Claude Opus 5.5
DeepSeek V3 is 96.5% cheaper for input tokens ($0.14 vs. $4.00 per 1M tokens) and $0.28 vs. $20.00 for output tokens (28.6x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (270ms faster than Claude Opus 5.5) with 78 vs. 95 tok/s throughput. Claude Opus 5.5 leads coding evaluations at 89.9% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 93.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 3.8 Flash
DeepSeek V3 is 44.0% cheaper for input tokens ($0.14 vs. $0.25 per 1M tokens) and $0.28 vs. $1.50 for output tokens (1.8x cost advantage). In terms of operational performance, Gemini 3.8 Flash delivers lower response latency at 105ms TTFT (305ms faster than DeepSeek V3) with 210 vs. 78 tok/s throughput. Gemini 3.8 Flash leads coding evaluations at 68.5% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 79.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GPT-5
DeepSeek V3 is 94.4% cheaper for input tokens ($0.14 vs. $2.50 per 1M tokens) and $0.28 vs. $10.00 for output tokens (17.9x cost advantage). In terms of operational performance, GPT-5 delivers lower response latency at 340ms TTFT (70ms faster than DeepSeek V3) with 105 vs. 78 tok/s throughput. GPT-5 leads coding evaluations at 74.9% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 85.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GPT-6 Astra
DeepSeek V3 is 98.6% cheaper for input tokens ($0.14 vs. $10.00 per 1M tokens) and $0.28 vs. $50.00 for output tokens (71.4x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (330ms faster than GPT-6 Astra) with 78 vs. 74 tok/s throughput. GPT-6 Astra leads coding evaluations at 75.7% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 96.0% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs DeepSeek V4 Pro
DeepSeek V3 is 89.4% cheaper for input tokens ($0.14 vs. $1.32 per 1M tokens) and $0.28 vs. $3.96 for output tokens (9.4x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (110ms faster than DeepSeek V4 Pro) with 78 vs. 78 tok/s throughput. DeepSeek V4 Pro leads coding evaluations at 80.6% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 90.1% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs DeepSeek R1
DeepSeek V3 is 74.5% cheaper for input tokens ($0.14 vs. $0.55 per 1M tokens) and $0.28 vs. $2.19 for output tokens (3.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (1040ms faster than DeepSeek R1) with 78 vs. 58 tok/s throughput. DeepSeek R1 leads coding evaluations at 49.2% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 71.5% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs DeepSeek V4 Flash Vision (Exp)
DeepSeek V3 is 36.4% cheaper for input tokens ($0.14 vs. $0.22 per 1M tokens) and $0.28 vs. $0.65 for output tokens (1.6x cost advantage). In terms of operational performance, DeepSeek V4 Flash Vision (Exp) delivers lower response latency at 310ms TTFT (100ms faster than DeepSeek V3) with 165 vs. 78 tok/s throughput. DeepSeek V4 Flash Vision (Exp) leads coding evaluations at 77.4% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 91.3% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs OpenAI o3
DeepSeek V3 is 93.0% cheaper for input tokens ($0.14 vs. $2.00 per 1M tokens) and $0.28 vs. $8.00 for output tokens (14.3x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (740ms faster than OpenAI o3) with 78 vs. 82 tok/s throughput. OpenAI o3 leads coding evaluations at 71.7% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 87.7% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Claude 4 Opus
DeepSeek V3 is 99.1% cheaper for input tokens ($0.14 vs. $15.00 per 1M tokens) and $0.28 vs. $75.00 for output tokens (107.1x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (510ms faster than Claude 4 Opus) with 78 vs. 46 tok/s throughput. Claude 4 Opus leads coding evaluations at 77.2% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 87.1% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Claude 3.7 Sonnet
DeepSeek V3 is 95.3% cheaper for input tokens ($0.14 vs. $3.00 per 1M tokens) and $0.28 vs. $15.00 for output tokens (21.4x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (130ms faster than Claude 3.7 Sonnet) with 78 vs. 72 tok/s throughput. Claude 3.7 Sonnet leads coding evaluations at 70.3% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 78.5% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 3.1 Pro
DeepSeek V3 is 88.8% cheaper for input tokens ($0.14 vs. $1.25 per 1M tokens) and $0.28 vs. $10.00 for output tokens (8.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (10ms faster than Gemini 3.1 Pro) with 78 vs. 98 tok/s throughput. Gemini 3.1 Pro leads coding evaluations at 71.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 82.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Grok 4
DeepSeek V3 is 94.4% cheaper for input tokens ($0.14 vs. $2.50 per 1M tokens) and $0.28 vs. $12.50 for output tokens (17.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (70ms faster than Grok 4) with 78 vs. 95 tok/s throughput. Grok 4 leads coding evaluations at 73.2% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 86.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Llama 4 Maverick
DeepSeek V3 is 30.0% cheaper for input tokens ($0.14 vs. $0.20 per 1M tokens) and $0.28 vs. $0.60 for output tokens (1.4x cost advantage). In terms of operational performance, Llama 4 Maverick delivers lower response latency at 135ms TTFT (275ms faster than DeepSeek V3) with 175 vs. 78 tok/s throughput. Llama 4 Maverick leads coding evaluations at 62.4% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 73.8% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 3.5 Flash-Lite
Gemini 3.5 Flash-Lite is 46.4% cheaper for input tokens ($0.07 vs. $0.14 per 1M tokens) and $0.30 vs. $0.28 for output tokens (1.9x cost advantage). In terms of operational performance, Gemini 3.5 Flash-Lite delivers lower response latency at 78ms TTFT (332ms faster than DeepSeek V3) with 265 vs. 78 tok/s throughput. Gemini 3.5 Flash-Lite leads coding evaluations at 55.4% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 68.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs OpenAI o4-mini
DeepSeek V3 is 87.3% cheaper for input tokens ($0.14 vs. $1.10 per 1M tokens) and $0.28 vs. $4.40 for output tokens (7.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (270ms faster than OpenAI o4-mini) with 78 vs. 125 tok/s throughput. OpenAI o4-mini leads coding evaluations at 68.1% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 81.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemma 4 31B IT
Gemma 4 31B IT is 14.3% cheaper for input tokens ($0.12 vs. $0.14 per 1M tokens) and $0.36 vs. $0.28 for output tokens (1.2x cost advantage). In terms of operational performance, Gemma 4 31B IT delivers lower response latency at 115ms TTFT (295ms faster than DeepSeek V3) with 155 vs. 78 tok/s throughput. Gemma 4 31B IT leads coding evaluations at 59.2% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 71.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs OpenAI o3-mini
DeepSeek V3 is 87.3% cheaper for input tokens ($0.14 vs. $1.10 per 1M tokens) and $0.28 vs. $4.40 for output tokens (7.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (540ms faster than OpenAI o3-mini) with 78 vs. 98 tok/s throughput. OpenAI o3-mini leads coding evaluations at 49.3% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 79.7% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Claude 3.5 Haiku
DeepSeek V3 is 82.5% cheaper for input tokens ($0.14 vs. $0.80 per 1M tokens) and $0.28 vs. $4.00 for output tokens (5.7x cost advantage). In terms of operational performance, Claude 3.5 Haiku delivers lower response latency at 190ms TTFT (220ms faster than DeepSeek V3) with 130 vs. 78 tok/s throughput. DeepSeek V3 leads coding evaluations at 42.0% SWE-bench Verified vs. 40.6% for Claude 3.5 Haiku and 59.1% vs. 41.6% on GPQA Diamond.
DeepSeek V3 vs Claude 3.5 Sonnet
DeepSeek V3 is 95.3% cheaper for input tokens ($0.14 vs. $3.00 per 1M tokens) and $0.28 vs. $15.00 for output tokens (21.4x cost advantage). In terms of operational performance, Claude 3.5 Sonnet delivers lower response latency at 310ms TTFT (100ms faster than DeepSeek V3) with 80 vs. 78 tok/s throughput. Claude 3.5 Sonnet leads coding evaluations at 49.0% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 59.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 2.0 Flash
Gemini 2.0 Flash is 28.6% cheaper for input tokens ($0.10 vs. $0.14 per 1M tokens) and $0.40 vs. $0.28 for output tokens (1.4x cost advantage). In terms of operational performance, Gemini 2.0 Flash delivers lower response latency at 280ms TTFT (130ms faster than DeepSeek V3) with 185 vs. 78 tok/s throughput. Gemini 2.0 Flash leads coding evaluations at 51.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 63.6% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 2.5 Flash
DeepSeek V3 is 53.3% cheaper for input tokens ($0.14 vs. $0.30 per 1M tokens) and $0.28 vs. $2.50 for output tokens (2.1x cost advantage). In terms of operational performance, Gemini 2.5 Flash delivers lower response latency at 110ms TTFT (300ms faster than DeepSeek V3) with 180 vs. 78 tok/s throughput. Gemini 2.5 Flash leads coding evaluations at 58.0% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 70.5% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Gemini 2.5 Pro
DeepSeek V3 is 88.8% cheaper for input tokens ($0.14 vs. $1.25 per 1M tokens) and $0.28 vs. $10.00 for output tokens (8.9x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (290ms faster than Gemini 2.5 Pro) with 78 vs. 80 tok/s throughput. Gemini 2.5 Pro leads coding evaluations at 63.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 74.0% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GLM-5.3
DeepSeek V3 is 60.0% cheaper for input tokens ($0.14 vs. $0.35 per 1M tokens) and $0.28 vs. $1.00 for output tokens (2.5x cost advantage). In terms of operational performance, GLM-5.3 delivers lower response latency at 160ms TTFT (250ms faster than DeepSeek V3) with 120 vs. 78 tok/s throughput. GLM-5.3 leads coding evaluations at 56.0% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 67.0% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GLM-5.3-Flash
GLM-5.3-Flash is 14.3% cheaper for input tokens ($0.12 vs. $0.14 per 1M tokens) and $0.40 vs. $0.28 for output tokens (1.2x cost advantage). In terms of operational performance, GLM-5.3-Flash delivers lower response latency at 90ms TTFT (320ms faster than DeepSeek V3) with 175 vs. 78 tok/s throughput. GLM-5.3-Flash leads coding evaluations at 52.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 63.5% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GPT-4o
DeepSeek V3 is 94.4% cheaper for input tokens ($0.14 vs. $2.50 per 1M tokens) and $0.28 vs. $10.00 for output tokens (17.9x cost advantage). In terms of operational performance, GPT-4o delivers lower response latency at 280ms TTFT (130ms faster than DeepSeek V3) with 100 vs. 78 tok/s throughput. GPT-4o leads coding evaluations at 48.0% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 53.6% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs GPT-4o Mini
DeepSeek V3 is 6.7% cheaper for input tokens ($0.14 vs. $0.15 per 1M tokens) and $0.28 vs. $0.60 for output tokens (1.1x cost advantage). In terms of operational performance, GPT-4o Mini delivers lower response latency at 220ms TTFT (190ms faster than DeepSeek V3) with 140 vs. 78 tok/s throughput. DeepSeek V3 leads coding evaluations at 42.0% SWE-bench Verified vs. 41.0% for GPT-4o Mini and 59.1% vs. 40.2% on GPQA Diamond.
DeepSeek V3 vs GPT-6.1 Sol
DeepSeek V3 is 93.0% cheaper for input tokens ($0.14 vs. $2.00 per 1M tokens) and $0.28 vs. $10.00 for output tokens (14.3x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (110ms faster than GPT-6.1 Sol) with 78 vs. 135 tok/s throughput. GPT-6.1 Sol leads coding evaluations at 86.4% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 92.4% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Grok 3
DeepSeek V3 is 95.3% cheaper for input tokens ($0.14 vs. $3.00 per 1M tokens) and $0.28 vs. $15.00 for output tokens (21.4x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (210ms faster than Grok 3) with 78 vs. 88 tok/s throughput. Grok 3 leads coding evaluations at 65.8% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 84.6% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Grok 3 Mini
DeepSeek V3 is 53.3% cheaper for input tokens ($0.14 vs. $0.30 per 1M tokens) and $0.28 vs. $1.20 for output tokens (2.1x cost advantage). In terms of operational performance, Grok 3 Mini delivers lower response latency at 150ms TTFT (260ms faster than DeepSeek V3) with 160 vs. 78 tok/s throughput. Grok 3 Mini leads coding evaluations at 54.0% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 68.0% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Llama 3.3 70B Instruct
Llama 3.3 70B Instruct is 14.3% cheaper for input tokens ($0.12 vs. $0.14 per 1M tokens) and $0.30 vs. $0.28 for output tokens (1.2x cost advantage). In terms of operational performance, Llama 3.3 70B Instruct delivers lower response latency at 145ms TTFT (265ms faster than DeepSeek V3) with 165 vs. 78 tok/s throughput. Llama 3.3 70B Instruct leads coding evaluations at 42.5% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 55.6% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Mistral Large 2 (2411)
DeepSeek V3 is 93.0% cheaper for input tokens ($0.14 vs. $2.00 per 1M tokens) and $0.28 vs. $6.00 for output tokens (14.3x cost advantage). In terms of operational performance, Mistral Large 2 (2411) delivers lower response latency at 380ms TTFT (30ms faster than DeepSeek V3) with 75 vs. 78 tok/s throughput. Mistral Large 2 (2411) leads coding evaluations at 45.2% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 54.2% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs o1
DeepSeek V3 is 99.1% cheaper for input tokens ($0.14 vs. $15.00 per 1M tokens) and $0.28 vs. $60.00 for output tokens (107.1x cost advantage). In terms of operational performance, DeepSeek V3 delivers lower response latency at 410ms TTFT (2190ms faster than o1) with 78 vs. 48 tok/s throughput. o1 leads coding evaluations at 48.9% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 78.0% vs. 59.1% on GPQA Diamond.
DeepSeek V3 vs Qwen 3.8 27B
DeepSeek V3 is 30.0% cheaper for input tokens ($0.14 vs. $0.20 per 1M tokens) and $0.28 vs. $0.60 for output tokens (1.4x cost advantage). In terms of operational performance, Qwen 3.8 27B delivers lower response latency at 120ms TTFT (290ms faster than DeepSeek V3) with 140 vs. 78 tok/s throughput. Qwen 3.8 27B leads coding evaluations at 58.4% SWE-bench Verified vs. 42.0% for DeepSeek V3 and 68.0% vs. 59.1% on GPQA Diamond.
Frequently Asked Questions & Query Fan-Out
How much does DeepSeek V3 cost per 1M tokens including prompt caching?
DeepSeek V3 costs $0.14 per million input tokens and $0.28 per million output tokens. Cached input reads cost $0.014 per 1M tokens, reducing repeated context costs by 90%.
What are DeepSeek V3's coding and reasoning benchmark scores?
DeepSeek V3 achieves 42% on SWE-bench Verified, 89.1% on HumanEval coding, 59.1% on GPQA Diamond graduate-level science, 75.9% on MMLU-Pro, and 90.2% on MATH-500, earning an overall Inference Value Ratio (IVR) of 75/100.
What is the context window, TTFT latency, and token throughput of DeepSeek V3?
DeepSeek V3 supports a 131,072 token context window (8,192 max output tokens) and operates at 410ms average Time to First Token (TTFT) with 78 tokens/sec output streaming throughput.
What are the primary production use cases for DeepSeek V3?
Cost-sensitive batch text processing, drafting, and open-weight self-hosted inference.