AI Model Profile
DeepSeek V4 Flash
The original DeepSeek V4 Flash family, with a retained July 31 quantization comparison.
Before you buy
Buying details reviewed 2026-10-22. Price, included amounts and access are dated observations; check the official source for current terms. This date does not review the whole fact ledger or renew historical demonstrations.
Best for: Operators evaluating the exact older Flash checkpoint on substantial local hardware.
Price: Canada/CAD: C$0 MIT weights; hardware and inference separate
Included: Downloadable weights are a separate route from today’s mutable hosted Flash API alias.
Main limits: The IQ2/IQ3 test used a six-A40 host. It does not qualify 128GB Macs, other quantizations, long context or the October 17 V4.1 update.
What Kingy showed: The retained 0731 comparison found only a two-point difference out of 426 between IQ2 and IQ3, while IQ2 used 13.281GB less payload.
Check official plans · See the demonstration and disclosure · Compare these workflows
- Family
- DeepSeek V4
- Release date
- Unknown
- Status
- Historical
- Context window
- 1,048,576 tokens: max_position_embeddings in the historical DeepSeek-V4-Flash-0731 checkpoint config. Current hosted V4.1 Flash documentation separately lists 1M context.
- Output limit
- Unknown
- API
- yes
- Open weights
- yes
- Local/self-hosted
- yes
- Pricing
- Usage-based DeepSeek API pricing; verify the official DeepSeek pricing page before budgeting.
- Evidence state
- Recheck due
Benchmark Caveat
Provider documentation and provider-published capability notes are directional, not universal rankings. Real results depend on prompts, tools, latency targets, pricing tier, safety filters, context length, and workload mix.
Best for
Operators evaluating the exact older Flash checkpoint on substantial local hardware.
Skip if
Skip if your team needs independently benchmarked performance claims or contractual guarantees not present in the official provider documentation.
Strengths
Official docs list a long context window, tool calls, JSON output, and low listed API prices.
Weaknesses
This profile does not independently validate benchmark quality, reliability, or latency.
Agent suitability
Candidate for agent tools because DeepSeek docs mention agent integrations and tool calls.
Kingy AI take
Use this profile as a source-backed reference point, not a ranking. Re-check official provider docs before making production or budget decisions.
Kingy AI Product Facts
DeepSeek V4 Flash
Current statusUnknown
- Company
- DeepSeek
- Primary job
- Not yet reviewed
- Audience
- Not yet reviewed
- API
- Available
- Evidence
- Source-backed; review due
- Latest source check
- November 3, 2026
See all tracked product factsPricing, platforms, dependencies, data claims, regions, and timeline
Sources and checks
- Coverage
- 10 of 88 tracked fields currently verified
- Sources
- 9
Source-check dates apply to the tracked facts. Editorial updates elsewhere on this page may have a different date.
Plans and pricing units
Not yet structured in this Kingy record.
Where it runs
- Availability
- Not yet reviewed
Developer access
- API access
- AvailableSource: Your First API Call · Using the Anthropic API · Using the Anthropic API · Using the Anthropic API · Your First API Call · Your First API Call · Your First API Call · Models & Pricing · Using the Anthropic API · Your First API Call · Your First API Call · Using the Anthropic API
- Protocol
- Anthropic API formatSource: Using the Anthropic API · Your First API Call · Using the Anthropic API
- API access
- AvailableSource: Your First API Call · Your First API Call · Your First API Call · Your First API Call · Your First API Call · Your First API Call · Your First API Call · Models & Pricing · Your First API Call · Your First API Call · Your First API Call
Stack and integrations
- Model / provider dependencies
- Not yet reviewed
- Integrations
- Not yet reviewed
Data and regions
- Vendor data-use claim
- Not yet reviewed
- Vendor retention claim
- Not yet reviewed
- Regions
- Not yet reviewed
Launch and latest material update
- Launch
- Not yet reviewed
- Latest material update
- Not yet reviewed
Full Model Notes
Which plan is worth considering
Canada/CAD: C$0 MIT weights; hardware and inference separate. Downloadable weights are a separate route from today’s mutable hosted Flash API alias.
Limits that affect the purchase
The IQ2/IQ3 test used a six-A40 host. It does not qualify 128GB Macs, other quantizations, long context or the October 17 V4.1 update. MIT weights carry no model license fee. Compatible compute, storage and serving still cost money. A local inference server exposes its own API; the current native deepseek-flash alias is not a version-pinned 0731 endpoint.
What the existing demonstration shows
The retained 0731 comparison found only a two-point difference out of 426 between IQ2 and IQ3, while IQ2 used 13.281GB less payload. Only IQ2 and IQ3 have complete scored runs. Q8 was a load-only smoke; long-context retrieval was not run. This describes the dated retained Kingy record, not a new product test. See the retained Kingy demonstration.
Sources and buying-review date
Buying details checked October 22, 2026. Plan research uses current official sources; demonstration descriptions use the retained Kingy record and any captured creator chapters. No new subscription, paid generation or checkout was performed. The buying-review date does not renew every claim in the broader facts ledger.
Check official plans · Compare related workflows
The Kingy Brief
Get the free AI price sheet.
25 AI plans and models on one page, from official pricing pages, updated every month. Free when you join The Kingy Brief: AI changes, original tests and one practical thing to try, Fridays at 09:00 Vancouver time.
Free · Double opt-in · Unsubscribe anytime
Signup help and newsletter schedule
After submitting, check your inbox for "Confirm your subscription to The Kingy Brief" and open its confirmation link. Your price sheet download is on the page that opens. Check Spam or Promotions if you cannot find it.
Fridays at 09:00 Vancouver time: source-checked AI changes, original tests and one practical thing to try. We skip a week when there is not enough verified material. Free. Unsubscribe anytime.
Coding notes
Evaluate coding workflows directly, especially if replacing other API backends.
Reasoning notes
Official docs describe thinking and non-thinking mode support.
Creative notes
Not primarily a creative/media model in this profile.
Research notes
Candidate for long-context and low-cost research tests with current API docs.
API pricing notes
Current hosted DeepSeek V4.1 Flash (deepseek-flash), including temporary deepseek-v4-flash compatibility routing: off-peak US$0.003 cache-hit input/US$0.15 cache-miss input/US$0.60 output per million tokens; peak US$0.006/US$0.30/US$1.20. Peak is Monday–Friday 01:00–04:00 and 06:00–10:00 UTC excluding Chinese public holidays; all other hours, weekends and Chinese public holidays are off-peak. Official documentation reviewed October 5, 2026; no paid endpoint test.
License notes
DeepSeek's official DeepSeek-V4-Flash-0731 model card states that the checkpoint repository and model weights are MIT-licensed. The separate hosted DeepSeek V4.1 Flash API route remains subject to DeepSeek's platform terms.
Official Model Links
Kingy coverage
Kingy AI articles that cover DeepSeek V4 Flash, newest first.
- BlogQwen3.8-27B vs DeepSeek V4 Flash-0731: Real Coding, Agent and Local-Use Tests
- BlogQwen3.8-Max vs Kimi K3 vs DeepSeek V4 Flash: Pricing, Coding, Context and Open Weights
- NewsDeepSeek V4-Flash Turns Better Training Into a Brutal AI Price Weapon
- BlogDeepSeek V4 Flash in Codex: 20 Real Coding Tasks, Cost, and Failure Rate
- BlogDeepSeek V4 Flash on 128GB Mac: 96.8GB Quant
- GuideDeepSeek V4 Flash on DGX Spark: 284B on a Desk, Kimi K3 in a Rack
- GuideCan Kimi K3 and DeepSeek V4 Flash Really Use 1 Million Tokens? The 1M Context Reality
- GuideKimi K3 vs. DeepSeek V4 Flash 0731: Which One Is Better?
Model Intelligence Research Map
Use these internal paths to move from this model profile into provider pages, static comparison pages, related Kingy records, and the broader AI launch graph. These are research paths, not rankings.