DeepSeek V4.5 Benchmarks, Pricing & Rankings
DeepSeek · 2026-06-14 · Open · Current
Snapshot 2026-10-14 · updated weekly
DeepSeek V4.5 ranks #21 of 42 on the AMR leaderboard.
DeepSeek V4.5 benchmark scores
| GPQA Diamond | 83.7% | |
| MMLU-Pro | 81.2% | |
| SWE-Bench Verified | 63.4% | |
| AIME 2025 | 85.1% | |
| LiveCodeBench v6 | 71.8% | |
| Arena Elo | 1385 |
DeepSeek V4.5 price, speed & context
Input price
$0.30
per 1M tokens
Output price
$1.20
per 1M tokens
Speed
112
tok/s
Context
256K
tokens
About DeepSeek V4.5
DeepSeek V4.5 set the open-source price floor for frontier-adjacent quality in mid-2026 and still anchors many budget inference stacks.
Models around DeepSeek V4.5 in the rankings
More DeepSeek models
DeepSeek V4.5 FAQ
Where does DeepSeek V4.5 rank?
DeepSeek V4.5 ranks #21 of 42 tracked AI models on the AMR aggregate as of October 2026, with an aggregate score of 84.5. Its strongest dimension is AIME 2025 at 85.1%.
How much does DeepSeek V4.5 cost?
DeepSeek V4.5 costs $0.30 per million input tokens and $1.20 per million output tokens on the standard API. At a typical 1:3 input/output agent workload, that blends to roughly $0.97 per million tokens.
What is the DeepSeek V4.5 context window?
DeepSeek V4.5 supports a 256K token context window and generates at a median 112 output tokens per second. That context fits roughly 171 standard documents in a single prompt.
Is DeepSeek V4.5 open source?
Yes — DeepSeek V4.5 ships open weights that you can download and self-host or serve through any compatible inference provider.
What is better than DeepSeek V4.5?
Muse Wonder ranks one place higher at #20 with an AMR score of 85.2. DeepSeek V4.5 still wins 2 of the tracked dimensions head-to-head — check the comparison table to see exactly which.