KymaKyma
DeepSeek

DeepSeek

DeepSeek V4 Pro

Top reasoning

DeepSeek's 1.6T-parameter MoE flagship, in Kyma's frontier-open quality tier — the model to reach for when reasoning depth matters more than speed or price.

Modalities

Text → Text

Below release

None reported

Weights

Published

deepseek-ai/DeepSeek-V4-Pro

Where it sits in the catalogue Kyma measures

Every number against every text model Kyma prices per token — a stated rule, not a chosen line-up.

Price · input#32/70

$0.6901

output $1.38 · cached $0.133

median $0.977 · best $0.0494

Price · output#20/70

$1.38

per 1M tokens generated

median $3.4515 · best $0.2164

Availability · 30d#9/70

99.02%

914 observations since 2026-05-02

median 94.0% · best 100.0%

Throughput#48/70

32.1 tok/s

probe median, one fixed prompt

median 39.8 tok/s · best 311.2 tok/s

Response time#51/70

3.48 s

probe median, to a complete answer

median 2.46 s · best 0.80 s

Context

1M

max output 65,536

published by DeepSeek, not measured here

Tick above each rail is this model, below it the other 69. Dashed rule is the field median, solid is its best. Better is left; the axis stops at the 90th percentile, so a few models sit past its right edge.

Usage

How much this model is actually called here.

Rank

#11

of 117 active models

Tokens served

79.2M

all-time

Platform share

2.3%

of all tokens

Tokens · last 15 daysAug 25 → Sep 8

How it behaves under real clients

The same model answering different prompt shapes — measured, not benchmarked.

ClientRequestsTokensThroughputTo first tokenp95 totalCompleted
OpenAI Python2132.8M39 tok/s4.18 s182 s100%
Copilot16713.1M33 tok/s3.63 s42 s100%

Grouped by the client that sent the request. Not a benchmark: the same model on different prompt shapes. Throughput and time-to-first-token are medians; p95 is the slowest response in twenty. 2 clients under 100 requests not shown.

Two clocks, and why they disagree

Kyma measures this model twice. Both are real; they answer different questions.

Probe · every 6h · 30 days

3.48s to answer

One fixed prompt, on a schedule, to every model. Comparable, because the model is the only thing that changes.

Observations
914
Answered by a substitute
3

Real traffic · last 7 days

15.07s to answer

Your requests, at the lengths clients actually send. Not comparable between models, but it is what running this one feels like.

Requests
232
Completed
96.6%
p95
144.7 s

Caching, as realised · last 7 days

39.0% of input cached

The share of input that actually hit cache, so the effective rate below is what was charged, not a best case.

Cached input
1.1M
Fresh input
1.7M
List input
$0.6901 /1M
Effective input
$0.472715 /1M

The gap is prompt length, not the model degrading. Use the probe figure to choose between models, the traffic figure to budget for your own.

Pricing

Pay per token. Cached input bills at this model’s own cached rate, listed below.

$0.6901 /1M input$1.38 /1M output
Hobby10 req/day · 2K in / 500 out
~$0.62/mo
Production1,000 req/day · 2K in / 500 out
~$62.11/mo
Scale20,000 req/day · 2K in / 500 out
~$1,242/mo
+ Estimate your workload
1,000
2,000
500
30%

Estimated monthly cost

$52.08

$1.74 / day on DeepSeek V4 Pro

Same workload on:

DeepSeek V4 Flash Vision$31.18-40%
DeepSeek V4 Flash$12.50-76%

Estimates use list pricing with cached input at this model's own cached rate. Actual bills depend on real token counts, and every response includes its exact cost.

When to use DeepSeek V4 Pro

Updated 2026-06-10

Where this model earns its cost — and where it doesn't.

DeepSeek V4 Pro is DeepSeek's flagship: a 1.6T mixture-of-experts model built for top-tier reasoning, complex coding, and long-context work. It sits in Kyma's frontier-open quality tier at premium pricing — the option you pick when the task is hard enough that output quality pays for itself.

On Kyma its traffic is led by coding agents. Every request gets automatic failover, and prompt caching is fully supported: repeated prompt prefixes bill at this model's cached input rate, published beside its input rate tier, where agents resend long prompts on every turn.

The 1M-token context window pairs with a 64K max output, function calling, structured outputs, and extended reasoning — headroom for very long inputs and long, structured generations.

Hard reasoning problems

A top-reasoning-tier model with extended reasoning, built for the math, planning, and analysis tasks where depth matters more than turnaround time.

Agentic coding

Coding agents already run production traffic through it on Kyma: OpenClaw is its top app by tokens, with Claude Code and Cline also in its top five.

Long-context work

The 1M-token window handles long documents, large specs, and large codebases in a single request.

Long structured generations

A 64K output ceiling plus structured outputs supports big refactors, detailed reports, and large JSON payloads.

Tool-driven pipelines

Function calling and structured outputs keep multi-step agent loops reliable when each step's correctness compounds.

Not ideal for: Image inputs (it is text-only) or latency-sensitive, high-volume workloads — it is a medium-speed, premium-priced model, so bulk background tasks run faster and cheaper on a fast-tier model.

How it compares

Against the peers people actually weigh it against.

SpecDeepSeek V4 ProDeepSeek V4 Flash VisionDeepSeek V4 Flash
Input /1M$0.6901$0.297$0.1389
Output /1M$1.38$0.891$0.2778
Context1M1M1M
ToolsYesYesYes
ReasoningYesYesYes
Throughput32.1 tok/s101.4 tok/s27.8 tok/s

Quick start

Up and running in under two minutes.

  1. 1

    Create an API key

    Sign up and grab a key from the dashboard. DeepSeek V4 Pro needs a top-up — the signup credit covers the free tier.

    Get API key →
  2. 2

    Make your first request

    Drop in your key and send a chat completion — fully OpenAI-compatible.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
      {"role": "user", "content": "Explain prompt caching in one paragraph."}
    ]
  }'
  3. 3

    Stream responses

    Add "stream": true to receive tokens as they arrive.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "stream": true,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

FAQ

Common questions about this model.

What is the context window of DeepSeek V4 Pro?

DeepSeek V4 Pro has a 1M-token context window — roughly 1471 pages of text in a single request.

How much does the DeepSeek V4 Pro API cost?

$0.6901 per 1M input tokens and $1.38 per 1M output tokens, with cached input at $0.133/1M — 81% off the input rate on repeated prompt prefixes. No subscription; you pay only for what you use.

Does DeepSeek V4 Pro support function calling?

Yes — DeepSeek V4 Pro supports tool/function calling and structured outputs (JSON mode), so it works with agent frameworks out of the box.

Are the weights for DeepSeek V4 Pro publicly available?

Yes. DeepSeek publishes DeepSeek V4 Pro's weights as deepseek-ai/DeepSeek-V4-Pro, so you can download and run the model yourself (https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro, read 2026-08-14). Kyma serves it because it is convenient and has failover behind it, not because it is the only way to reach it.

Is DeepSeek V4 Pro ever served below the precision its creator released it at?

Not as far as Kyma can see. DeepSeek V4 Pro was released by DeepSeek at fp4 + fp8 (https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro, read 2026-08-14), and no route serving it here reports a format below that, as checked on 2026-09-03. Read that as the absence of a claim rather than as a guarantee: a route publishes its precision only where it chooses to, and most state nothing either way.

How do I use DeepSeek V4 Pro?

Kyma is OpenAI-compatible: point your SDK's base URL at https://kymaapi.com/v1, use your Kyma API key, and set the model to deepseek-v4-pro. Signing up is free, but this model needs a top-up: the $0.50 signup credit covers the free tier only.

Is DeepSeek V4 Pro good for coding agents?

Yes — coding agents are a major share of its Kyma traffic. OpenClaw is its top app by tokens, with Claude Code and Cline also in its top five, and its combination of extended reasoning, function calling, and a 64K output ceiling suits long, multi-step code changes. For quick edits or high-volume background tasks, a cheaper, faster model usually makes more economic sense.

How does DeepSeek V4 Pro perform in production on Kyma?

It is a medium-speed model — pick it for output quality rather than latency, and lean on prompt caching to keep repeated prefixes at this model's cached input rate.

Why run DeepSeek V4 Pro through Kyma?

One OpenAI-compatible endpoint and one key covers this and every other model on the platform. You get automatic failover when a serving path degrades, exact per-request cost in usage.cost, and prompt caching that bills repeated prefixes at this model's cached input rate, which is published beside its input rate.

Needs a top-up — the signup credit covers the free tier.Create account →

More models by DeepSeek

ModelContextInputOutput
DeepSeekDeepSeek V4 Flash Vision1M$0.297$0.891
DeepSeekDeepSeek V4 Flash1M$0.1389$0.2778
DeepSeekDeepSeek V3160K$0.351$0.513
DeepSeekDeepSeek R164K$0.7425$2.957