Coding Horizon

DeepSeek V4 Flash Makes Claude Look Expensive Now

Every figure, price, date, version and licence this video puts on screen, chased to a primary source. Checked 22 August 2026.

DeepSeek V4 Flash, the model

Claim Value Source
Total parameters 284B Hugging Face, DeepSeek V4 Flash model card
Active parameters per token 13B Hugging Face model card
Architecture Mixture of experts Hugging Face model card
Context length 1,000,000 tokens DeepSeek API docs, pricing
Maximum output 384,000 tokens DeepSeek API docs, pricing
Thinking and non thinking modes Both, thinking is the default DeepSeek API docs, pricing
Licence MIT Hugging Face model card
Open weights Published on Hugging Face Hugging Face model card
Current build DeepSeek V4 Flash 0731 DeepSeek API docs, pricing
Concurrency limit 2,500 DeepSeek API docs, pricing

DeepSeek V4 Pro, the model

Claim Value Source
Total parameters 1.6T DeepSeek API docs, V4 release notes
Active parameters per token 49B DeepSeek API docs, V4 release notes
Context length 1,000,000 tokens DeepSeek API docs, pricing
Thinking and non thinking modes Both DeepSeek API docs, pricing
Current build DeepSeek V4 Pro 0813 DeepSeek API docs, pricing
Concurrency limit 500 DeepSeek API docs, pricing

Both V4 models were announced together, both open weight, both at one million context.

DeepSeek API pricing

DeepSeek moved from flat rates to peak and off peak billing on 16 August 2026. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, and off peak rates are half the peak rates. Per one million tokens, in US dollars:

Model Cache hit input Cache miss input Output
V4 Flash, off peak $0.007 $0.22 $0.66
V4 Flash, peak $0.014 $0.44 $1.32
V4 Pro, off peak $0.022 $0.66 $1.98
V4 Pro, peak $0.044 $1.32 $3.96

Source: DeepSeek API docs, pricing.

The same page notes a further change to the peak and off peak rules effective 23 August 2026, applying off peak rates throughout the day at weekends.

Caching is automatic disk based prefix caching: a request whose prompt shares a prefix with an earlier one bills the shared portion at the cache hit rate.

The narration quotes the flat rates that applied before 16 August 2026 ($0.14 cache miss, $0.0028 cache hit and $0.28 output for Flash; $0.435, $0.003625 and $0.87 for Pro). Those were correct until that date and are no longer DeepSeek’s published prices. The script was supplied finished and is recorded word for word, so the narration could not be corrected. No DeepSeek price appears as a figure anywhere in the picture. The pricing shots draw the shape of the gap instead: relative tier heights, model marks at the height their price puts them, and multiples between tiers. See ### Not checked in MANIFEST.md.

Premium routes, for comparison

These are current and are printed on screen.

Model Input per 1M Output per 1M Source
Claude Opus 4.6 $5 $25 Claude platform docs, pricing
Claude Sonnet 5 $2 $10 Claude platform docs, pricing
GPT 5 $1.25 $10 OpenAI, API pricing

Anthropic’s cache read price is 0.1 times the base input price, and a five minute cache write is 1.25 times base, so Opus 4.6 reads a cache hit at $0.50 per million and Sonnet 5 at $0.20 per million.

Claude Sonnet 5 pricing correction. The narration says Sonnet 5 is $2 and $10 through 31 August 2026 and then moves to $3 and $15. Anthropic’s pricing page now states that the $2 and $10 rate, announced at launch as introductory pricing through 31 August 2026, is now the standard price, and that the scheduled increase to $3 and $15 on 1 September 2026 will not occur. The picture shows $2 and $10 as the standard price and never shows $3 or $15.

Caveats