DeepSeek V4.5

DeepSeekreasoning NewOpen Source

DeepSeek's July 5 2026 open-weight release under an MIT license — the strongest open reasoning model on price-per-quality. SWE-bench Pro 62.1%, GPQA Diamond 91.8%, and competition-math scores that sit with the proprietary frontier. Hosted API pricing is $0.50/$1.10 per 1M tokens; weights are downloadable for on-prem or private-cloud inference, so the marginal token cost self-hosted is infrastructure only. Benchmarks were self-reported at launch; the technical report and full eval harness followed within a week.

Context Window

256K

tokens

Max Output

33K

tokens

Input Price

$0.5

per 1M tokens

Output Price

$1.1

per 1M tokens

Speed

62

tokens/sec

Released

Jul 2026

2026-07-05

Blended Cost

$0.80

per 1M tokens

Value Score

115.0

quality per $

Capabilities

ChatFunction CallingCode GenerationReasoning

Benchmarks

Quality Index
92
MMLU Pro
89.4
HumanEval (Coding)
91.2
MATH
93.1
Arena ELO
1471

Open Source — Licensed under MIT

About DeepSeek V4.5

DeepSeek V4.5 is a reasoning AI model by DeepSeek, released on July 5, 2026. It supports a context window of 256K tokens and can generate up to 33K output tokens.

At $0.5 per million input tokens and $1.1 per million output tokens, its blended cost of $0.80/1M tokens makes it one of the most affordable models available. Its value score of 115.0 reflects the balance of quality and cost.

DeepSeek V4.5 is available as an open-source model under the MIT license, meaning you can self-host it for predictable costs or use it through API providers like Swfte Connect.

Using DeepSeek V4.5 with Swfte

Access DeepSeek V4.5 through Swfte Connect, our unified LLM gateway. Connect gives you a single API for 50+ models, with automatic routing, cost optimization, and fallback handling. You can also try DeepSeek V4.5 in our AI Playground before integrating.