Model Class

DeepSeek V4

DeepSeek's flagship V3 successor, served as V4-Pro (1M context) and V4-Flash, whose 75% price cut became the permanent list price on May 22, 2026. Per [DeepSeek V4-Pro 75% Price Cut Goes Permanent](https://codersera.com/blog/deepseek-v4-pro-permanent-price-cut-may-2026/).

8 connections3 resources5 posts

Summary

What it is

DeepSeek's flagship V3 successor, served as V4-Pro (1M context) and V4-Flash, whose 75% price cut became the permanent list price on May 22, 2026. Per [DeepSeek V4-Pro 75% Price Cut Goes Permanent](https://codersera.com/blog/deepseek-v4-pro-permanent-price-cut-may-2026/).

Where it fits

It is a model-class node on the cost-of-inference axis of the site. Its permanent price floor ($0.435/$0.87 per MTok for Pro) makes token-heavy patterns — agentic retrieval, validators, large-batch data processing on S3-backed corpora — economically viable, and it sets a reference price competitors must answer.

Misconceptions / Traps
  • The flagship API name is V4-Pro, not bare "V4"; V4-Flash is a separate, cheaper tier — don't conflate their prices.
  • "Permanent" is the publisher's framing for a non-expiring list price, not a contractual guarantee; verify live rates before quoting.
  • The cache-hit price ($0.003625/M Pro, $0.0028/M Flash) only applies to cached input, not all input — blended cost depends on cache-hit ratio.
  • Competitor figures (GPT-5.5, Opus 4.7) come from third-party analysis, not DeepSeek's docs; treat multipliers as reported, not official.
Key Connections
  • extends DeepSeek V3 — V4 is the direct successor flagship line.
  • solves High Cloud Inference Cost — its permanent floor is the clearest 2026 datapoint on collapsing inference cost.
  • optimizes_for Performance-per-Dollar — positioned explicitly on price/capability against US frontier models.
  • competes_with GPT-5.5 / Claude Opus — third-party pricing puts it ~11-34x cheaper per token.

Definition

What it is

DeepSeek V4 is DeepSeek's flagship model family and the successor to V3, served via the first-party API in two tiers: V4-Pro (1M context, thinking and non-thinking modes) and the cheaper V4-Flash. On May 22, 2026 DeepSeek made its 75% V4-Pro discount permanent — converting a promotion that was set to expire May 31, 2026 into the standing list price. Per DeepSeek V4-Pro 75% Price Cut Goes Permanent.

Why it exists

Cheap, capable inference is the economic precondition for agent-heavy data infrastructure — agentic retrieval (DCI), validators, and LLM-assisted catalog/data systems all burn tokens. By publishing a non-expiring price floor, V4 directly drives down the cost-of-inference for self-hosted-adjacent AI-data pipelines. Per DeepSeek's steep V4-Pro price cut escalates AI pricing war (InfoWorld).

Primary use cases

Cost-sensitive high-volume inference, agentic retrieval and search loops, batch document/data processing, long-context (1M) reasoning, LLM-assisted data and catalog tooling.

Connections8

Outbound7
scoped_to1
AI Runtime Infrastructure
extends1
DeepSeek V3
implements2
DeepSeekMoEMulti-Head Latent Attention (MLA)
optimizes_for1
Performance-per-Dollar
solves1
High Cloud Inference Cost
competes_with1
Claude Fable 5
Inbound1
competes_with1
Claude Fable 5

Resources3

DocsHigh
api-docs.deepseek.com/quick_start/pricing

Authoritative first-party pricing for V4-Pro and V4-Flash (input cache-hit/miss and output).

BlogHigh
www.infoworld.com/article/4176709/deepseeks-steep-v4-pro-pri...

Reputable trade-press confirmation of the permanent framing and pricing-war context.

BlogMedium
codersera.com/blog/deepseek-v4-pro-permanent-price-cut-may-2...

Provides the May 22, 2026 date and the head-to-head multipliers vs GPT-5.5 and Claude Opus 4.7.

Featured in

2026-09-03When the Quota Became a VariableWe measured our own $20 inference plan dropping from ~2M tokens per session to ~700K — same plan, same price, no changelog. Then we pulled the industry-wide evidence: a documented 2025–2026 wave of quota re-denominations, peak-hour throttles admitted only after user telemetry caught them, and deprecations that silently rebill at 6× the rate. Also the counter-evidence, because the thesis only survives if it's tested: OpenAI uncapped text, Anthropic doubled limits in May, and DeepSeek's price floor keeps falling. The pattern isn't uniform gouging — it's bifurcation plus unit games. And it's the strongest argument yet for the stack this index maps.2026-07-05The Frontier Moved Again — and the Floor CompoundedIn June we argued the AI stack had split into a closed, expensive frontier and an open, cheap floor. Six weeks later the frontier proved the point the hard way: Claude Fable 5 was pulled offline for 19 days by an export-control order, then restored with a retention mandate that quietly bars the most sensitive data from ever touching it. Meanwhile the floor didn't just hold — it compounded. Object storage got RDMA-fast, the query engines converged on one substrate, and the data plane grew a control plane built for agents instead of humans. This is the field report.2026-06-16The Whole Stack Went Open — Weights, Storage, and SovereigntyIn the same quarter that a 1.6-trillion-parameter open-weights model landed on top of last year's closed frontier — and Anthropic's Claude Fable 5 promptly sprinted ahead again — the storage layer underneath it went open too: DeepSeek open-sourced its file system, Europe stood up sovereign S3, and the post-MinIO self-hosted stack matured. The frontier and the floor are pulling apart, and the pattern underneath all of it is the oldest pain point this index maps: vendor lock-in.2026-06-16The Frontier and the Floor: The AI Stack Just Split in TwoIn a single quarter, Anthropic's Claude Fable 5 reset the closed frontier while DeepSeek V4 put frontier-of-last-year capability into open weights at roughly one-fiftieth the price. The two events look like a race. They're the opposite: the AI stack is bifurcating into a closed, expensive frontier for the hardest autonomous work and an open, cheap floor for the high-volume inference that actually runs your data infrastructure. The question stopped being 'which model' and became 'which tier.'2026-06-06Storage Grew an Agent InterfaceThe S3 API made storage programmable for applications. In a six-week stretch of 2026, the storage layer grew a second interface — for agents. Weaviate, Pinecone, and RustFS all shipped built-in MCP servers; cheap inference and agentic retrieval made it inevitable; and the Apache Polaris CVE cluster showed the security bill that comes due. This is the discoverability inflection for self-hosted AI data infrastructure.