# DeepSeek V4.1 Flash Turns Agent Memory Into a Pricing Weapon
**Plutonous** | September 12, 2026 | 8 min read
> DeepSeek-V4.1-Flash pairs 1M-token context with a $0.006-per-million peak cache-hit price. Explore the architecture, API changes and economics of reusable agent context.
Tags: DeepSeek, AI Agents, Inference Economics, Open Weights, KV Cache, Model APIs, Long Context, AI Infrastructure
---
*Cover: AI-generated editorial artwork illustrating asymmetric input and output stages. The blocks are a conceptual metaphor.*
**TL;DR: DeepSeek-V4.1-Flash lists a $0.006-per-million-token peak cache-hit input price, against $0.30 for cache-miss input and $1.20 for output. On a fixed illustrative invoice of 1 million billed input tokens and 100,000 output tokens, a 90% input-cache hit lowers the charge from $0.4200 to $0.1554, a 63.0% reduction.[3] That is why the model's 1M-token context and claimed 890 bytes per token global KV cache matter commercially, even though neither figure proves a faster or cheaper completed agent task.[4]**
DeepSeek released DeepSeek-V4.1-Flash on September 10 as an MIT-licensed, open-weight multimodal mixture-of-experts model. The company describes a **552B-parameter backbone** with only **8B active parameters for input/prefill** and **16B for output/decode**, plus a 1M-token context window and 384K maximum output on its hosted API.[1][3][4]
[External or live embed omitted from static article text.]
[View DeepSeek’s original launch post on X](https://x.com/deepseek_ai/status/2097930608790167907).
The real story isn't the 552B number. It is that DeepSeek is trying to make persistent context cheap enough that agents can keep rereading repositories, tool histories, policies, and case files without turning every loop into a premium inference event.
### Why This Matters Now
The agent market is becoming a memory market. A model's input list price matters less when a long-running workflow can reliably reuse its prefix, and more when cache misses, retries, tool output, and verification force the system to rebuild context. DeepSeek's listed API economics make cache match rate a product metric, not a serving footnote.[3]
## The Pricing: Cache Hits Change the Unit of Competition
DeepSeek lists peak API prices of **$0.006/M** cache-hit input, **$0.30/M** cache-miss input, and **$1.20/M** output for `deepseek-flash`. Its listed off-peak prices are exactly half: $0.003, $0.15, and $0.60 per million tokens. Peak windows are 01:00–04:00 and 06:00–10:00 UTC on weekdays.[3] Those are hosted-API list prices, not a self-hosting cost estimate or a promise of future pricing.
Here is the genius: the 50-fold spread between cache-hit and cache-miss input makes application architecture visible in the invoice. An agent that carries a stable system prompt, repository snapshot, and policy bundle can make the expensive material reusable. An agent that constantly changes prompt order, formatting, or tenant-specific prefixes can reduce cache hits. DeepSeek is effectively charging developers to learn cache discipline.
DeepSeek’s caching guide makes the condition concrete: a request must fully match a prefix unit already persisted by the service. Caching is automatic and best-effort; a repeated document does not guarantee an immediate hit. Developers can measure the result through `prompt_cache_hit_tokens` and `prompt_cache_miss_tokens` in API usage data.[13]
### A 90% Input Cache Hit Cuts This Fixed Invoice by 63.0%
Illustrative invoice
Same 1,000,000 billed input tokens and 100,000 output tokens at DeepSeek's listed peak API prices. The only changed assumption is the share of billed input that matches cache.
- label: No input cache match; value: 42; displayValue: $0.4200; description: $0.3000 input + $0.1200 output
- label: 90% input cache match; value: 15.54; displayValue: $0.1554; description: $0.0054 cached input + $0.0300 miss input + $0.1200 output
42
USD per illustrative billed workload
$0
$0.4200
Source: DeepSeek Models & Pricing. Assumes cache match, fixed billed tokens, and peak pricing; excludes retries and does not measure task cost, latency, quality, or savings realized in production.
The uncomfortable truth is that a cache-hit price is only useful when the workload earns it. A coding agent that changes the repository state, inserts fresh tool traces, or relies on uncached retrieval may consume far more miss-priced input than its happy-path demo suggests. Output is also unchanged by the cache assumption. In this example, the $0.1200 output bill becomes most of the remaining cost.
## The Architecture: Asymmetric Activation Is the Commercial Design
DeepSeek calls V4.1-Flash a Causal Encoder-Decoder design. The model card says it uses 20 causal encoder layers and 20 decoder layers; its architecture activates 8B parameters per prefill token and 16B per decode token across the model's input-and-output path.[4] This is an architecture description, not a published end-to-end throughput result.
That serving focus follows [DeepSeek’s earlier DeepSpec work on speculative decoding](/news/deepseek-deepspec-speculative-decoding-inference-economics). V4.1 adds a different pressure point: the memory used to carry context forward.
The company also says the global KV cache uses **890 bytes per token**, roughly one quarter of its V4-Flash predecessor, while persistent SSD cache needs roughly one eighth as much. Read that precisely: 890 bytes/token describes the model's **global KV cache**, not total VRAM, full model memory, a minimum server configuration, or the cost of an agent deployment.[4]
The technical report says its SWA Bounded Replay approximately reconstructs short-window states instead of persisting them, trading a small amount of prefill recomputation for lower storage. DeepSeek reports negligible quality degradation; that remains a vendor characterization of a storage-computation trade-off, not a claim of lossless replay.[10]
That distinction matters. DeepSeek's reference inference guide demonstrates checkpoint conversion with eight-way tensor parallelism (`MP=8`) and explicitly presents itself as a readable reference implementation. It does not publish a minimum VRAM requirement, validated minimum server configuration or measured cost per completed agent task.[6] Open weights expand control. They do not make a 552B-backbone system a workstation model.
### The Published Shape of V4.1 Flash
DeepSeek's technical and API disclosures define capabilities and price signals. They do not define a reproducible serving bill of materials.
- label: Backbone parameters; value: 552B; description: DeepSeek's model-card architecture figure, distinct from a deployment memory estimate.
- label: Active on input; value: 8B; description: Vendor-described prefill activation in the causal encoder.
- label: Active on output; value: 16B; description: Vendor-described active parameters across the model during decode.
- label: Global KV cache; value: 890 bytes/token; description: Vendor-reported cache footprint, not total VRAM or total serving memory.
- label: Hosted context window; value: 1M; description: Listed API context ceiling; it is not a throughput or latency benchmark.
Sources: DeepSeek model card, technical report, API pricing documentation, and reference inference README. Architecture and cache figures are vendor-reported.
## The Product Migration: An Alias Is Not a Stable Model Contract
DeepSeek says `deepseek-flash` is the current API model ID. Its older `deepseek-v4-flash` and `deepseek-v4-flash-vision-exp` names are accepted as compatibility aliases and served by V4.1-Flash at Flash pricing.[2] That removes migration friction, but it also means callers can receive a materially different model without changing a string in their code.
The company initially announced that `deepseek-v4-pro` traffic would route to V4.1-Flash from September 14 at 04:00 UTC. Its current changelog reverses that position: V4-Pro will continue after September 14 with unchanged billing, citing user demand.[1][2] The latest operational statement wins. Production teams should treat the Pro reroute as withdrawn, not as an imminent migration.
The operational conclusion is straightforward: pin evaluations to the endpoint behavior that matters, keep regression suites, and monitor alias changes as closely as price changes. A stable API string is not a stable behavioral contract when it is an actively managed compatibility alias.
SGLang and Miles announced day-zero inference and reinforcement-learning support, respectively, for V4.1-Flash. That widens the integration path for builders, but it is support availability, not evidence of a particular latency, throughput, or cost outcome.[12]
## The Benchmarks: A Profile, Not a Speed Crown
DeepSeek reports **90.6** on Terminal-Bench 2.1 under its disclosed agent protocol: reasoning effort 100, temperature 1.0, top-p 0.95, DSH Minimal mode, 1M context, three runs, Linux containers, 500 maximum steps, and no network.[5] That is useful evidence of what DeepSeek tested. It is vendor-reported, and the public Harbor listing was imported from the model card rather than produced as an independent V4.1 evaluation.[7][8]
Artificial Analysis assigns V4.1-Flash a **40** on its Intelligence Index for the reasoning model at maximum effort. Its model page says the suite consumed **250 million output tokens** in aggregate and flags unusually high verbosity. That is independent measurement, but it is an aggregate suite result, not one request's token use or a cost-per-task estimate.[9]
[External or live embed omitted from static article text.]
[View Artificial Analysis’s original post on X](https://x.com/ArtificialAnlys/status/2098148674203488422).
Let's be clear: these evaluations are not normalized speed tests. They do not establish a measured self-hosted cost per completed task, and a tool-using, max-effort agent score should not become a model-only ranking across different scaffolds.
What buyers should compare is accepted-result cost: the cost to get a correct, verified patch, report, or workflow completion after retries and human review. Cache reuse can lower the input part of that equation dramatically. It cannot compensate for an agent that takes bad actions, generates excessive output, or fails on the final verification step.
### What the Announcement Establishes, and What It Does Not
- Business implication
- Still unproven
- feature: 1M context and $0.006/M cache-hit input; values: - Large stable prefixes can be inexpensive on the hosted API
- Cache-match rate, retries, and accepted-result cost
- feature: 552B backbone; 8B prefill / 16B decode active; values: - DeepSeek is optimizing different input and output phases
- Tok/s, TTFT, tail latency, and required serving hardware
- feature: MIT weights and TP8 reference conversion; values: - Buyers have an exit and integrators can build support
- Minimum VRAM, efficient cluster topology, and self-hosted TCO
- feature: 90.6 Terminal-Bench 2.1 vendor report; values: - A specified evaluation profile exists
- Independent model-only superiority or a direct speed ranking
## The Verdict: DeepSeek Wants Memory to Become a Margin Lever
DeepSeek-V4.1-Flash turns a familiar systems question, whether an agent can reuse its working context, into a sharply differentiated price signal. At peak list rates, 900,000 cache-hit input tokens cost $0.0054; the same number of cache-miss tokens cost $0.2700.[3]
A larger context window is only a capacity feature until a product can afford to use it repeatedly. DeepSeek is making that repeated use central to its pricing. The strategic wager is that agent builders will standardize prefixes, preserve histories, and organize tool flows around reusable state. If they do, cache efficiency becomes part of their gross margin. If they do not, the attractive headline price is mostly a theoretical discount.
/images/articles/deepseek-v41-flash/cache-reuse.webp
Editorial illustration of a compact black workspace encircled by an organized loop of black and cream archive modules on a rail.
Generated editorial artwork illustrating the reuse of organized context around a compact active workspace. It is a conceptual visualization, not a technical diagram or performance measurement.
1672
941
16/9
cover
940
### The Key Due-Diligence Question
Do not buy a cache-hit price. Test a representative agent trace: prompt construction, cache-match rate, tool output growth, retries, verification, latency at concurrency, and accepted-result quality. DeepSeek's published prices and cache architecture describe an opportunity. They do not measure your finished-work cost.
DeepSeek has supplied a credible mechanism and an aggressive price card. The market now has to prove whether its agent stacks can preserve enough reusable context to turn that mechanism into margin. That is a harder contest than winning a launch-day benchmark table, and it is the one that matters.
## Sources
1. [DeepSeek-V4.1-Flash Release Announcement](https://www.deepseek.com/en/news/deepseek-v4-1-flash/)
2. [DeepSeek API Changelog](https://api-docs.deepseek.com/updates/)
3. [DeepSeek Models & Pricing](https://api-docs.deepseek.com/quick_start/pricing)
4. [DeepSeek-V4.1-Flash Model Card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash)
5. [Evaluation Results](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash#evaluation-results)
6. [Official Minimal Inference README](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/inference/README.md)
7. [Terminal-Bench 2.1 Dataset and Leaderboard](https://huggingface.co/datasets/harborframework/terminal-bench-2.1)
8. [Evaluation Result Import Discussion](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/discussions/29)
9. [DeepSeek-V4.1-Flash Model Analysis](https://artificialanalysis.ai/models/deepseek-v4-1-flash)
10. [DeepSeek-V4.1-Flash Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdf)
11. [DeepSeek V4.1-Flash Launch Post](https://x.com/deepseek_ai/status/2097930608790167907)
12. [SGLang and Miles Support for DeepSeek-V4.1-Flash](https://www.lmsys.org/blog/2026-09-10-deepseek-v41/)
13. [Context Caching](https://api-docs.deepseek.com/guides/kv_cache/)
*Last updated: September 12, 2026*
---
*Source: [LLM Rumors](https://www.llmrumors.com/news/deepseek-v41-flash-pricing-agent-memory)*