# DeepSeek V4 Flash

DeepSeek V4 Flash er en model fra DeepSeek med åbne vægte, først registreret den 2026-09-20; lanceringsdatoen er ikke fastlagt. Kontekstvinduet er 1.048.576 tokens.

- Udvikler: DeepSeek
- Id: deepseek-ai/DeepSeek-V4-Flash
- Ifølge rapporter: 2026-04-24T05:22:50.000Z
- Observation: 2026-09-20T14:43:15.580Z
  - [kildelink](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash)
- Status: Lanceret
- Vægtversion: 2026-04-23
  - [kildelink](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Flash/paths-info/60d8d70770c6776ff598c94bb586a859a38244f1)
- Kanonisk adresse: https://alexgetman.com/da/models/deepseek/deepseek-v4-flash-418/

## Spørgsmål

### Hvor stort er kontekstvinduet for DeepSeek V4 Flash?

Kontekstvinduet er 1.048.576 tokens.

### Er vægtene for DeepSeek V4 Flash offentligt tilgængelige?

Ja. DeepSeek offentliggør vægtene.

### Hvornår blev DeepSeek V4 Flash lanceret?

Lanceringsdatoen er ikke fastlagt. Signal Forge registrerede modellen første gang den 2026-09-20.

## Tidslinje

- 2026-07-31T11:26:38.354Z · [Alex Getman](https://alexgetman.com/137/yesterday-gpt-5-6-luna-pushed-the-pareto-frontier-for-intelligence-and-cost-toda/)
> Yesterday GPT 5.6 Luna pushed the Pareto frontier for intelligence and cost. Today DeepSeek V4 Flash did it again and pushed it even further.
- 2026-08-03T08:01:30.947Z · [Alex Getman](https://alexgetman.com/142/alibaba-released-qwen-3-8-max/)
> ⚡️ Alibaba released Qwen 3.8 Max
> 
> The new 2.4T parameter flagship benchmarks in the same league as GPT-5.6 Sol, Claude Opus 4.8 and Fable 5.
> 
> API pricing is $2/M input and $6/M output, roughly half the cost of GPT-5.6 Terra and Claude Sonnet.
> 
> Now it’s up to users to see whether Qwen can beat DeepSeek-V4-Flash and GPT-5.6 Luna on price to performance.
- 2026-08-17T07:00:00.000Z · [Alex Getman](https://alexgetman.com/192/deepseek-slashed-opencode-go-limits-17x/)
> 🚨 DeepSeek slashed OpenCode Go limits 17x
> 
> DeepSeek V4 Flash limits dropped from 63,300 to 3,800, while V4 Pro fell from 3,450 to 1,050. This comes as DeepSeek API prices have jumped sharply.
- 2026-11-05T17:52:07.101Z · [DeepSeek · updates](https://api-docs.deepseek.com/updates#deepseek-v4-flash-update)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · 0xPlaygrounds/rig · @gold-silver-copper](https://github.com/0xPlaygrounds/rig/pull/3298)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · BerriAI/litellm · @cloudgpuapp](https://github.com/BerriAI/litellm/pull/48331)
> Usage: `litellm.completion(model="cloudgpu/deepseek-v4-flash", messages=[...])`
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · I-GUIDE/iguide-ai · @ShanmukhUpad](https://github.com/I-GUIDE/iguide-ai/pull/184)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · NeuroLang/NeuroLang · @demianw](https://github.com/NeuroLang/NeuroLang/pull/1201)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · NeuroLang/NeuroLang · @demianw](https://github.com/NeuroLang/NeuroLang/pull/1202)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · NixOS/nixpkgs · @asosnovsky](https://github.com/NixOS/nixpkgs/pull/565378)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · Picazo333/nexova-ai-engineering · @Picazo333](https://github.com/Picazo333/nexova-ai-engineering/pull/18)
> Modelo de generación: `online/openrouter/deepseek/deepseek-v4-flash` en `https://llm.4geeks.ai/v1`. Embeddings: `online/openrouter/perplexity/pplx-embed-v1-0.6b`.
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · SreeSharvesh/loc-arena · @ryanlail](https://github.com/SreeSharvesh/loc-arena/pull/316)
> - `…control-note.glm-flash.yaml` and `…control-note.deepseek-v4-flash.yaml`: the same run with GLM 5.3 Flash and DeepSeek V4 Flash as the untrusted agent, each with its model file.
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · alibaba/rtp-llm · @fxp19950422](https://github.com/alibaba/rtp-llm/pull/1676)
> Brings the DeepSeek V4 Flash PPU decode path to the execution shape measured in #1434. Every new behavior is opt-in or keeps its previous default.
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · anomalyco/models.dev · @models-dev-automation[bot]](https://github.com/anomalyco/models.dev/pull/13568)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · gufo-org/gufo · @mduka](https://github.com/gufo-org/gufo/pull/1210)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · maximhq/bifrost · @neta79](https://github.com/maximhq/bifrost/pull/4376)
> "model": "deepseek/deepseek-v4-flash",
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · maximhq/bifrost · @neta79](https://github.com/maximhq/bifrost/pull/4468)
> -d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"Say exactly: hello world"}],"max_tokens":20}'
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · maximhq/bifrost · @Amit-kr26](https://github.com/maximhq/bifrost/pull/6809)
> [x] Live run with a real key against `deepseek-v4-flash`: simple chat, multi-turn, streaming, tool calls (single/multiple/automatic/end-to-end), list models (incl. pagination), Responses API (incl. streaming), and
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · maximhq/bifrost · @jimaldon](https://github.com/maximhq/bifrost/pull/7124)
> needed. The model catalog datasheet already carries `tencent/deepseek-v4-pro`, `tencent/deepseek-v4-flash` and `tencent/minimax-m3`; live-only models are listed with blank pricing and can be priced via overrides.
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · nikgraf/calendar · @nikgraf](https://github.com/nikgraf/calendar/pull/316)
- 2026-11-13T18:47:56.425Z · [GitHub search · DeepSeek · pulls · overnightworks/speech · @dependabot[bot]](https://github.com/overnightworks/speech/pull/354)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · ModernNomad-98/Project-Aegis · @deepseek-v4-flash](https://github.com/ModernNomad-98/Project-Aegis/commit/222e26158187b038b8255a288d76652a5f0ce205)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · ModernNomad-98/Project-Aegis · @deepseek-v4-flash](https://github.com/ModernNomad-98/Project-Aegis/commit/f0e3dac0a1e66daba4bc952217e2cbb75d2c4368)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · bottomless/agent-duel · @shihabcodes](https://github.com/bottomless/agent-duel/commit/148ea95c1252abe424dabbc68102cb174e1a2b45)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · genji1024/dotfiles · @genji1024](https://github.com/genji1024/dotfiles/commit/cbec5cd555d340f2290d8c9a1eceec6d8f17dc34)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · miron404/lindwyrm · @miron404](https://github.com/miron404/lindwyrm/commit/371592c516717417afc399a10bf88e993c5769ae)
> defaulted to the retired deepseek-v4-flash model name.
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · nikgraf/calendar · @nikgraf](https://github.com/nikgraf/calendar/commit/29cc5fdecaf4cd63b3d0c0a7b9ab299873b326fb)
- 2026-11-13T18:48:53.232Z · [GitHub search · DeepSeek · commits · rgolde96/natively-ai-personal · @unknown](https://github.com/rgolde96/natively-ai-personal/commit/96def36267b41ddbc3340df740b43d107e52f498)
> feat(llm): DeepSeek V4 Flash on OpenRouter with reasoning off and Fireworks/Wafer/DeepInfra order (D-30); record owner answers, WS23, owner info page
- 2026-11-15T10:37:24.000Z · [GitHub search · DeepSeek · pulls · faizan45640/PakCommerceAi · @faizan45640](https://github.com/faizan45640/PakCommerceAi/pull/97)
> Found in a live smoke test against DeepSeek: `deepseek-v4-flash` runs in thinking mode by default, which rejects the forced `tool_choice` used by structured output (HTTP
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · Finkregh/stepstone-hierarchy · @renovate[bot]](https://github.com/Finkregh/stepstone-hierarchy/pull/11)
> - Fixed DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level ([#​8181](https://redirect.github.com/earendil-works/pi/pull/8181) by [@​tianshuang](https://redirect.github.com/tianshuang)).
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · SillyTavern/SillyTavern · @Wolfsblvt](https://github.com/SillyTavern/SillyTavern/pull/6213)
> handling: #6180's pattern, which also covers the legacy `deepseek-v4-flash` id. DeepSeek still accepts that id and serves it with the current Flash model, and other
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39678)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39682)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · fitchmultz/pi-agent-skills · @fitchmultz](https://github.com/fitchmultz/pi-agent-skills/pull/125)
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · kingithegreat/Sadie · @kingithegreat](https://github.com/kingithegreat/Sadie/pull/500)
> by DeepSeek-V4.1-Flash-1017. The docs say the legacy name `deepseek-v4-flash` (and `deepseek-v4-flash-vision-exp`) is "still accepted, but the corresponding models have been retired". `deepseek-chat` and `deepseek-reasoner` had
- 2026-11-15T10:52:39.021Z · [GitHub search · DeepSeek · pulls · sgl-project/sglang · @AkiYangDev](https://github.com/sgl-project/sglang/pull/50517)
> common precomputed routing. The full-model checks below use DeepSeek-V4-Flash W8A8. V4.1 uses the same
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · ChidiRnweke/FollowThrough.ai · @ChidiRnweke](https://github.com/ChidiRnweke/FollowThrough.ai/pull/509)
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · HansHKApps/promptquorum-website · @HansHKApps](https://github.com/HansHKApps/promptquorum-website/pull/386)
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · dietrichliko/vfgh-llm-bench · @dietrichliko](https://github.com/dietrichliko/vfgh-llm-bench/pull/2)
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39679)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39680)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39681)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39685)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · onprem-ai/deployment-catalog · @tomas-polach](https://github.com/onprem-ai/deployment-catalog/pull/19)
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · openclaw/openclaw · @SlowikInspiruje](https://github.com/openclaw/openclaw/pull/194556)
> the comment at the top of the file. `deepseek-v4-flash` and `deepseek-v4-pro` are listed; `deepseek-flash` is the same contract under a later id and was simply
- 2026-11-15T11:10:24.173Z · [GitHub search · DeepSeek · pulls · wxm0394/Switchyard · @wxm0394](https://github.com/wxm0394/Switchyard/pull/23)
> the max_tokens=10 probes that succeeded were served by DeepSeek-V4-Flash with **1 completion token** after the reasoning models (Kilo Ultra, Lightning-30b, GLM-5.2) spent the whole budget
- 2026-11-15T11:11:15.219Z · [GitHub search · DeepSeek · commits · dlprevent-dev/dlprevent · @unknown](https://github.com/dlprevent-dev/dlprevent/commit/ee1dc07e3cb4e70979c32f360a3a24f41ce9c394)
> DeepSeek V4 Flash thinks by default. Its reasoning comes back in
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · br41s/hermes-sandbox · @br41s](https://github.com/br41s/hermes-sandbox/pull/136)
> hermes -p bl-shoroban config set model.default deepseek/deepseek-v4-flash hermes -p bl-shoroban config set model.provider openrouter
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · br41s/hermes-sandbox · @br41s](https://github.com/br41s/hermes-sandbox/pull/218)
> chore(models): update deepseek-v4-flash refs to v4.1-flash
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · cheaper-inference/OmniRoute · @binyangzhu000-sudo](https://github.com/cheaper-inference/OmniRoute/pull/16114)
> Anthropic, OpenAI, …) under vendor-prefixed ids such as `deepseek-ai/deepseek-v4-flash`.
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · cheaper-inference/OmniRoute · @tonydavis629](https://github.com/cheaper-inference/OmniRoute/pull/18463)
> fresh pay-as-you-go key: non-streamed and streamed chat on `deepseek/deepseek-v4-flash` with `reasoning_effort: high` both returned 200 with the expected text, and a function tool on `z-ai/glm-5.3`
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39677)
> node ~/.agents/skills/contribute-to-eliza/scripts/run-receipt.mjs disclose --provider deepseek --model deepseek-v4-flash --client hermes-agent
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · next-moe/nextmoe-infra · @KunMoe](https://github.com/next-moe/nextmoe-infra/pull/528)
> tag and trait ids in browse_works and browse_characters deepseek-v4-flash sends `"tag_ids": [84]` (JSON numbers) even though the schema declares string items. `BrowseWorks` / `BrowseCharacters` decoded
- 2026-11-15T11:28:29.553Z · [GitHub search · DeepSeek · pulls · tinygrad/tinygrad · @b1tg](https://github.com/tinygrad/tinygrad/pull/19303)
> llm: deepseek-v4-flash
- 2026-11-15T11:44:45.925Z · [GitHub search · DeepSeek · pulls · ID-Robots/clawbox · @KrasimirKralev](https://github.com/ID-Robots/clawbox/pull/1502)
> | Chat model, provider and `clawaiConfigured` unchanged (`deepseek-v4-flash` on `clawai`) | ok | ok |
- 2026-11-15T11:44:45.925Z · [GitHub search · DeepSeek · pulls · Kreijstal/pi-managed-workflow · @renovate[bot]](https://github.com/Kreijstal/pi-managed-workflow/pull/3)
> - Fixed DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level ([#​8181](https://redirect.github.com/earendil-works/pi/pull/8181) by [@​tianshuang](https://redirect.github.com/tianshuang)).
- 2026-11-15T11:44:45.925Z · [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/731)
- 2026-11-15T11:44:45.925Z · [GitHub search · DeepSeek · pulls · octos-org/octos · @ymote](https://github.com/octos-org/octos/pull/3245)
> serve --shared --solo` from this branch, real model `deepseek-v4-flash`:
- 2026-11-15T11:44:45.925Z · [GitHub search · DeepSeek · pulls · wyrd-company/ahp-pi-agent · @renovate[bot]](https://github.com/wyrd-company/ahp-pi-agent/pull/4)
> - Fixed DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level ([#​8181](https://redirect.github.com/earendil-works/pi/pull/8181) by [@​tianshuang](https://redirect.github.com/tianshuang)).
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · WolfMark-Ltd/spendsense · @WolfMarkTools](https://github.com/WolfMark-Ltd/spendsense/pull/61)
> | `deepseek-v4-flash (nan) list price changed on models.dev.` | `nan now lists deepseek-v4-flash input at $0.29484 per 1M tokens (previously $0; no meaningful percentage)` |
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39669)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · flashinfer-ai/flashinfer · @YijiaZhao](https://github.com/flashinfer-ai/flashinfer/pull/8806)
> is built for decode-size batches. In an earlier DeepSeek-V4-Flash sweep on RTX 6000D with the three backends on identical routes, it was faster up to
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · lidge-jun/opencodex · @henrikdukefoss](https://github.com/lidge-jun/opencodex/pull/8444)
> - **Seeded models.** `StandardCompute` (the default, routed), `deepseek/deepseek-v4-pro`, `deepseek/deepseek-v4-flash`, `z-ai/glm-5.3` and `moonshotai/kimi-k3`, taken from the live catalog on 2026-11-09. The live list replaces the seed
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · metadist/synaplan · @orgaralf](https://github.com/metadist/synaplan/pull/3122)
- 2026-11-15T12:05:07.628Z · [GitHub search · DeepSeek · pulls · ml-explore/mlx-swift-lm · @Augustas11](https://github.com/ml-explore/mlx-swift-lm/pull/796)
> Adds the `deepseek_v4` text model, the architecture of DeepSeek-V4-Flash. This is the DeepSeek-V4 model PR I listed in #221 on
- 2026-11-15T12:07:53.867Z · [GitHub search · DeepSeek · commits · ai-dynamo/aisim · @YijiaZhao](https://github.com/ai-dynamo/aisim/commit/ea6bb6d95685d47fa01d856cb8e8e7b02b82713d)
> (DeepSeek-V4-Flash, AgentX): on TP8 at concurrency 2 and 8 the model no
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · CaffeeLake/Roo-Cline · @renovate[bot]](https://github.com/CaffeeLake/Roo-Cline/pull/20)
> - [`00bfad5`](https://redirect.github.com/vercel/ai/commit/00bfad5): feat(provider/deepseek): add DeepSeek V4 Flash and Pro model IDs
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · OptimNow/cloud-finops-skills · @OptimNow](https://github.com/OptimNow/cloud-finops-skills/pull/374)
> | 2026-11-06 | Databricks: GLM 5.2 retires 4 February 2027; DeepSeek V4 Flash (0731) retired 5 November | https://docs.databricks.com/aws/en/machine-learning/retired-models-policy | medium |
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · RBLN-SW/vllm-rbln · @rebel-mhkang](https://github.com/RBLN-SW/vllm-rbln/pull/1457)
> model: add DeepSeek-V4 Flash, PP and MTP support
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · boettiger-lab/open-llm-proxy · @cboettig](https://github.com/boettiger-lab/open-llm-proxy/pull/191)
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · btraceio/jafar · @jbachorik](https://github.com/btraceio/jafar/pull/205)
> — the test reads the developer's real `~/.config/jafar/llm.properties` (`llm.model=deepseek-v4-flash:cloud`); not hermetic on machines with that file. Untouched by this diff.
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · cloudsnuchoi/2026NASA · @cloudsnuchoi](https://github.com/cloudsnuchoi/2026NASA/pull/32)
> api-docs.deepseek.com의 Models & Pricing 페이지에서 확인했다. 레거시 이름 `deepseek-v4-flash`는 쓰지 않는다.
- 2026-11-15T12:23:13.343Z · [GitHub search · DeepSeek · pulls · jpenilla/pi-cloudflare-ai-gateway-native · @renovate[bot]](https://github.com/jpenilla/pi-cloudflare-ai-gateway-native/pull/10)
> - Fixed DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level ([#​8181](https://redirect.github.com/earendil-works/pi/pull/8181) by [@​tianshuang](https://redirect.github.com/tianshuang)).
- 2026-11-15T12:24:35.114Z · [GitHub search · DeepSeek · commits · hallchristine827/hal · @hallchristine827](https://github.com/hallchristine827/hal/commit/7004acf60f2cff7a5ee720969442498656582cdd)
> 本文提供一种基于 TaoToken 统一 API 通道反推 WorkBuddy credits 实际人民币价值的方法：通过固定提示词调用 DeepSeek V4 Flash/Pro，获取 token usage 并按官方 API 定价折算成本，再除以对应 credits 消耗，得出单 credit 单价（约 ¥0.032–0.034）。关键在于统一 Base URL、严格对齐请求参数，并规避多模型密钥管理复杂性，支撑套餐性价比判断。
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · BerriAI/litellm · @PCTEJA](https://github.com/BerriAI/litellm/pull/51879)
> 1. They call litellm.completion with model opencode_go/deepseek-v4-flash, reasoning_effort none, and drop_params enabled
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · Murathanx12/Aegis-Finance · @Murathanx12](https://github.com/Murathanx12/Aegis-Finance/pull/160)
> in + 40 out tokens per row on `deepseek-v4-flash` ≈ **USD 4.99 for the whole window, ~USD 0.055/day** (estimate; no call made). Given the audit,
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · anomalyco/models.dev · @models-dev-automation[bot]](https://github.com/anomalyco/models.dev/pull/13717)
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · ed-donner/undercover · @Feyisa-Diba](https://github.com/ed-donner/undercover/pull/144)
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39665)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39687)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · ggml-org/llama.cpp · @kh0pper](https://github.com/ggml-org/llama.cpp/pull/28571)
> - DeepSeek-V4-Flash Q4_K_XL two-box on a tree without the Vulkan fused HC ops: the ops resolve to the RPC device and the worker dies with
- 2026-11-15T12:40:40.023Z · [GitHub search · DeepSeek · pulls · wyolet/relay-catalog · @aaliboyev](https://github.com/wyolet/relay-catalog/pull/66)
> a default anchor policy, drops the draft's retired `deepseek-chat`/`deepseek-reasoner`/`deepseek-v4-flash` bindings and stale pricing. V4 Pro's release date is corrected to its GA (2026-08-13); legacy DeepSeek
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · NoFxAiOS/nofx · @deanokk](https://github.com/NoFxAiOS/nofx/pull/1488)
> Added `deepseek-v4-flash` and `deepseek-v4-pro` to the Claw402 backend route map, added both model IDs to the internal price table, and placed both models at the
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · NousResearch/hermes-agent · @teknium1](https://github.com/NousResearch/hermes-agent/pull/155648)
> Before / after (real `_model_listing_reply`, scratch home: `model.default: deepseek-v4-flash`, only `DEEPSEEK_API_KEY` set)
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39659)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39663)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39664)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39688)
> AI provider/model: deepseek / deepseek-v4-flash Client / agent tooling: hermes-agent
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · huynhtandat223/firstmate · @huynhtandat223](https://github.com/huynhtandat223/firstmate/pull/212)
> Flash`, and leaves no record or launch; MD2 `deepseek-v4-flash` passes. `tests/fm-control-relaunch.test.sh` exit 0 (incl. the relaunch case: Pro refuses before the agent stops with the
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/724)
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/729)
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · mposition/Tomverse · @mposition](https://github.com/mposition/Tomverse/pull/3976)
> by construction, but two cases name specific models (`deepseek-v4-flash` after Model Finder, `claude-haiku-4-5` selected) and need CI to confirm.
- 2026-11-15T12:55:48.073Z · [GitHub search · DeepSeek · pulls · scribekey-app/scribekey-models · @github-actions[bot]](https://github.com/scribekey-app/scribekey-models/pull/87)
> `{"input": 0.32, "output": 0.89}`; outputLimit: `16000` → `16384`); `deepseek/deepseek-v4-flash` (price: `{"input": 0.091, "output": 0.182}` → `{"input": 0.09, "output": 0.18}`; outputLimit: `943717` → `393216`); `deepseek/deepseek-v4-flash-0731` (price:
- 2026-11-15T12:56:56.389Z · [GitHub search · DeepSeek · commits · Pathoschild/StardewMalwareScanner · @Pathoschild](https://github.com/Pathoschild/StardewMalwareScanner/commit/9b62909691416af3102956b930fe5b1c17920cde)
> bump suggested AI model from DeepSeek V4 Flash to V4.1 Flash

## Kildelinks

- [editorial](https://alexgetman.com/137/yesterday-gpt-5-6-luna-pushed-the-pareto-frontier-for-intelligence-and-cost-toda/)
- [Alex Getman](https://alexgetman.com/142/alibaba-released-qwen-3-8-max/)
- [editorial](https://alexgetman.com/192/deepseek-slashed-opencode-go-limits-17x/)
- [DeepSeek · updates](https://api-docs.deepseek.com/updates#deepseek-v4-flash-update)
- [GitHub search · DeepSeek · pulls · 0xPlaygrounds/rig · @gold-silver-copper](https://github.com/0xPlaygrounds/rig/pull/3298)
- [GitHub search · DeepSeek · pulls · BerriAI/litellm · @cloudgpuapp](https://github.com/BerriAI/litellm/pull/48331)
- [GitHub search · DeepSeek · pulls · BerriAI/litellm · @PCTEJA](https://github.com/BerriAI/litellm/pull/51879)
- [GitHub search · DeepSeek · pulls · CaffeeLake/Roo-Cline · @renovate[bot]](https://github.com/CaffeeLake/Roo-Cline/pull/20)
- [GitHub search · DeepSeek · pulls · ChidiRnweke/FollowThrough.ai · @ChidiRnweke](https://github.com/ChidiRnweke/FollowThrough.ai/pull/509)
- [GitHub search · DeepSeek · pulls · Finkregh/stepstone-hierarchy · @renovate[bot]](https://github.com/Finkregh/stepstone-hierarchy/pull/11)
- [GitHub search · DeepSeek · pulls · HansHKApps/promptquorum-website · @HansHKApps](https://github.com/HansHKApps/promptquorum-website/pull/386)
- [GitHub search · DeepSeek · pulls · I-GUIDE/iguide-ai · @ShanmukhUpad](https://github.com/I-GUIDE/iguide-ai/pull/184)
- [GitHub search · DeepSeek · pulls · ID-Robots/clawbox · @KrasimirKralev](https://github.com/ID-Robots/clawbox/pull/1502)
- [GitHub search · DeepSeek · pulls · Kreijstal/pi-managed-workflow · @renovate[bot]](https://github.com/Kreijstal/pi-managed-workflow/pull/3)
- [GitHub search · DeepSeek · commits · ModernNomad-98/Project-Aegis · @deepseek-v4-flash](https://github.com/ModernNomad-98/Project-Aegis/commit/222e26158187b038b8255a288d76652a5f0ce205)
- [GitHub search · DeepSeek · commits · ModernNomad-98/Project-Aegis · @deepseek-v4-flash](https://github.com/ModernNomad-98/Project-Aegis/commit/f0e3dac0a1e66daba4bc952217e2cbb75d2c4368)
- [GitHub search · DeepSeek · pulls · Murathanx12/Aegis-Finance · @Murathanx12](https://github.com/Murathanx12/Aegis-Finance/pull/160)
- [GitHub search · DeepSeek · pulls · NeuroLang/NeuroLang · @demianw](https://github.com/NeuroLang/NeuroLang/pull/1201)
- [GitHub search · DeepSeek · pulls · NeuroLang/NeuroLang · @demianw](https://github.com/NeuroLang/NeuroLang/pull/1202)
- [GitHub search · DeepSeek · pulls · NixOS/nixpkgs · @asosnovsky](https://github.com/NixOS/nixpkgs/pull/565378)
- [GitHub search · DeepSeek · pulls · NoFxAiOS/nofx · @deanokk](https://github.com/NoFxAiOS/nofx/pull/1488)
- [GitHub search · DeepSeek · pulls · NousResearch/hermes-agent · @teknium1](https://github.com/NousResearch/hermes-agent/pull/155648)
- [GitHub search · DeepSeek · pulls · OptimNow/cloud-finops-skills · @OptimNow](https://github.com/OptimNow/cloud-finops-skills/pull/374)
- [GitHub search · DeepSeek · commits · Pathoschild/StardewMalwareScanner · @Pathoschild](https://github.com/Pathoschild/StardewMalwareScanner/commit/9b62909691416af3102956b930fe5b1c17920cde)
- [GitHub search · DeepSeek · pulls · Picazo333/nexova-ai-engineering · @Picazo333](https://github.com/Picazo333/nexova-ai-engineering/pull/18)
- [GitHub search · DeepSeek · pulls · RBLN-SW/vllm-rbln · @rebel-mhkang](https://github.com/RBLN-SW/vllm-rbln/pull/1457)
- [GitHub search · DeepSeek · pulls · SillyTavern/SillyTavern · @Wolfsblvt](https://github.com/SillyTavern/SillyTavern/pull/6213)
- [GitHub search · DeepSeek · pulls · SreeSharvesh/loc-arena · @ryanlail](https://github.com/SreeSharvesh/loc-arena/pull/316)
- [GitHub search · DeepSeek · pulls · WolfMark-Ltd/spendsense · @WolfMarkTools](https://github.com/WolfMark-Ltd/spendsense/pull/61)
- [GitHub search · DeepSeek · commits · ai-dynamo/aisim · @YijiaZhao](https://github.com/ai-dynamo/aisim/commit/ea6bb6d95685d47fa01d856cb8e8e7b02b82713d)
- [GitHub search · DeepSeek · pulls · alibaba/rtp-llm · @fxp19950422](https://github.com/alibaba/rtp-llm/pull/1676)
- [GitHub search · DeepSeek · pulls · anomalyco/models.dev · @models-dev-automation[bot]](https://github.com/anomalyco/models.dev/pull/13568)
- [GitHub search · DeepSeek · pulls · anomalyco/models.dev · @models-dev-automation[bot]](https://github.com/anomalyco/models.dev/pull/13717)
- [GitHub search · DeepSeek · pulls · boettiger-lab/open-llm-proxy · @cboettig](https://github.com/boettiger-lab/open-llm-proxy/pull/191)
- [GitHub search · DeepSeek · commits · bottomless/agent-duel · @shihabcodes](https://github.com/bottomless/agent-duel/commit/148ea95c1252abe424dabbc68102cb174e1a2b45)
- [GitHub search · DeepSeek · pulls · br41s/hermes-sandbox · @br41s](https://github.com/br41s/hermes-sandbox/pull/136)
- [GitHub search · DeepSeek · pulls · br41s/hermes-sandbox · @br41s](https://github.com/br41s/hermes-sandbox/pull/218)
- [GitHub search · DeepSeek · pulls · btraceio/jafar · @jbachorik](https://github.com/btraceio/jafar/pull/205)
- [GitHub search · DeepSeek · pulls · cheaper-inference/OmniRoute · @binyangzhu000-sudo](https://github.com/cheaper-inference/OmniRoute/pull/16114)
- [GitHub search · DeepSeek · pulls · cheaper-inference/OmniRoute · @tonydavis629](https://github.com/cheaper-inference/OmniRoute/pull/18463)
- [GitHub search · DeepSeek · pulls · cloudsnuchoi/2026NASA · @cloudsnuchoi](https://github.com/cloudsnuchoi/2026NASA/pull/32)
- [GitHub search · DeepSeek · pulls · dietrichliko/vfgh-llm-bench · @dietrichliko](https://github.com/dietrichliko/vfgh-llm-bench/pull/2)
- [GitHub search · DeepSeek · commits · dlprevent-dev/dlprevent · @unknown](https://github.com/dlprevent-dev/dlprevent/commit/ee1dc07e3cb4e70979c32f360a3a24f41ce9c394)
- [GitHub search · DeepSeek · pulls · ed-donner/undercover · @Feyisa-Diba](https://github.com/ed-donner/undercover/pull/144)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39659)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39663)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39664)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39665)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39669)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39677)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39678)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39679)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39680)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39681)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39682)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39685)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39687)
- [GitHub search · DeepSeek · pulls · elizaOS/eliza · @mtcow](https://github.com/elizaOS/eliza/pull/39688)
- [GitHub search · DeepSeek · pulls · faizan45640/PakCommerceAi · @faizan45640](https://github.com/faizan45640/PakCommerceAi/pull/97)
- [GitHub search · DeepSeek · pulls · fitchmultz/pi-agent-skills · @fitchmultz](https://github.com/fitchmultz/pi-agent-skills/pull/125)
- [GitHub search · DeepSeek · pulls · flashinfer-ai/flashinfer · @YijiaZhao](https://github.com/flashinfer-ai/flashinfer/pull/8806)
- [GitHub search · DeepSeek · commits · genji1024/dotfiles · @genji1024](https://github.com/genji1024/dotfiles/commit/cbec5cd555d340f2290d8c9a1eceec6d8f17dc34)
- [GitHub search · DeepSeek · pulls · ggml-org/llama.cpp · @kh0pper](https://github.com/ggml-org/llama.cpp/pull/28571)
- [GitHub search · DeepSeek · pulls · gufo-org/gufo · @mduka](https://github.com/gufo-org/gufo/pull/1210)
- [GitHub search · DeepSeek · commits · hallchristine827/hal · @hallchristine827](https://github.com/hallchristine827/hal/commit/7004acf60f2cff7a5ee720969442498656582cdd)
- [GitHub search · DeepSeek · pulls · huynhtandat223/firstmate · @huynhtandat223](https://github.com/huynhtandat223/firstmate/pull/212)
- [GitHub search · DeepSeek · pulls · jpenilla/pi-cloudflare-ai-gateway-native · @renovate[bot]](https://github.com/jpenilla/pi-cloudflare-ai-gateway-native/pull/10)
- [GitHub search · DeepSeek · pulls · kingithegreat/Sadie · @kingithegreat](https://github.com/kingithegreat/Sadie/pull/500)
- [GitHub search · DeepSeek · pulls · lidge-jun/opencodex · @henrikdukefoss](https://github.com/lidge-jun/opencodex/pull/8444)
- [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/724)
- [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/729)
- [GitHub search · DeepSeek · pulls · luongnv89/freetokens · @luongnv89](https://github.com/luongnv89/freetokens/pull/731)
- [GitHub search · DeepSeek · pulls · maximhq/bifrost · @neta79](https://github.com/maximhq/bifrost/pull/4376)
- [GitHub search · DeepSeek · pulls · maximhq/bifrost · @neta79](https://github.com/maximhq/bifrost/pull/4468)
- [GitHub search · DeepSeek · pulls · maximhq/bifrost · @Amit-kr26](https://github.com/maximhq/bifrost/pull/6809)
- [GitHub search · DeepSeek · pulls · maximhq/bifrost · @jimaldon](https://github.com/maximhq/bifrost/pull/7124)
- [GitHub search · DeepSeek · pulls · metadist/synaplan · @orgaralf](https://github.com/metadist/synaplan/pull/3122)
- [GitHub search · DeepSeek · commits · miron404/lindwyrm · @miron404](https://github.com/miron404/lindwyrm/commit/371592c516717417afc399a10bf88e993c5769ae)
- [GitHub search · DeepSeek · pulls · ml-explore/mlx-swift-lm · @Augustas11](https://github.com/ml-explore/mlx-swift-lm/pull/796)
- [GitHub search · DeepSeek · pulls · mposition/Tomverse · @mposition](https://github.com/mposition/Tomverse/pull/3976)
- [GitHub search · DeepSeek · pulls · next-moe/nextmoe-infra · @KunMoe](https://github.com/next-moe/nextmoe-infra/pull/528)
- [GitHub search · DeepSeek · commits · nikgraf/calendar · @nikgraf](https://github.com/nikgraf/calendar/commit/29cc5fdecaf4cd63b3d0c0a7b9ab299873b326fb)
- [GitHub search · DeepSeek · pulls · nikgraf/calendar · @nikgraf](https://github.com/nikgraf/calendar/pull/316)
- [GitHub search · DeepSeek · pulls · octos-org/octos · @ymote](https://github.com/octos-org/octos/pull/3245)
- [GitHub search · DeepSeek · pulls · onprem-ai/deployment-catalog · @tomas-polach](https://github.com/onprem-ai/deployment-catalog/pull/19)
- [GitHub search · DeepSeek · pulls · openclaw/openclaw · @SlowikInspiruje](https://github.com/openclaw/openclaw/pull/194556)
- [GitHub search · DeepSeek · pulls · overnightworks/speech · @dependabot[bot]](https://github.com/overnightworks/speech/pull/354)
- [GitHub search · DeepSeek · commits · rgolde96/natively-ai-personal · @unknown](https://github.com/rgolde96/natively-ai-personal/commit/96def36267b41ddbc3340df740b43d107e52f498)
- [GitHub search · DeepSeek · pulls · scribekey-app/scribekey-models · @github-actions[bot]](https://github.com/scribekey-app/scribekey-models/pull/87)
- [GitHub search · DeepSeek · pulls · sgl-project/sglang · @AkiYangDev](https://github.com/sgl-project/sglang/pull/50517)
- [GitHub search · DeepSeek · pulls · tinygrad/tinygrad · @b1tg](https://github.com/tinygrad/tinygrad/pull/19303)
- [GitHub search · DeepSeek · pulls · wxm0394/Switchyard · @wxm0394](https://github.com/wxm0394/Switchyard/pull/23)
- [GitHub search · DeepSeek · pulls · wyolet/relay-catalog · @aaliboyev](https://github.com/wyolet/relay-catalog/pull/66)
- [GitHub search · DeepSeek · pulls · wyrd-company/ahp-pi-agent · @renovate[bot]](https://github.com/wyrd-company/ahp-pi-agent/pull/4)
- [hf-author-artifacts:deepseek-ai](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Flash/paths-info/60d8d70770c6776ff598c94bb586a859a38244f1)
- [review](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash)

## Signaler