Comparison
DeepSeek V4 Flash & Pro vs GPT-5.4 vs Claude Opus 4.6: Official Pricing and Capability Comparison
EvoLink Team
Product Team
March 7, 2026
Updated on September 10, 2026
11 min read
Lifecycle update — September 10, 2026: DeepSeek released V4.1 Flash. On DeepSeek's direct API,deepseek-v4-flashanddeepseek-v4-flash-vision-expnow forward to V4.1 Flash, anddeepseek-v4-prois scheduled to follow on September 14, 2026 at 12:00 Beijing time (04:00 UTC). On EvoLink,deepseek-v4-flashanddeepseek-v4-proare not affected and continue to serve DeepSeek V4 Flash and V4 Pro;deepseek-v4-flash-vision-expnow redirects to DeepSeek V4.1 Flash. See the official update, the V4.1 Flash model page and the migration guide.
Updated August 13, 2026
This historical cross-provider comparison records the August 13 DeepSeek request IDs, Flash 0731 and Pro 0813 versions, and the direct-provider pricing checked at that time. The URL continues to own the Claude Opus 4.6 comparison intent; use the linked model pages for current route availability and billing.
Context Note
Claude Opus 4.7 launched on April 16, 2026 and is now generally available from Anthropic. This article intentionally keeps Claude Opus 4.6 in scope because the URL and query target are specifically about the
4.6 comparison set. Treat this page as a historical and still-useful comparison baseline rather than the final word on the latest Claude release.If you are comparing DeepSeek V4, GPT-5.4, and Claude Opus 4.6, the most important update is this: as of August 13, 2026, DeepSeek maps the request IDs
deepseek-v4-flash and deepseek-v4-pro to DeepSeek-V4-Flash-0731 and DeepSeek-V4-Pro-0813. Both list 1M context, 384K max output, tool calls, Responses API, and Anthropic API support. DeepSeek Models & PricingThat means the comparison has changed from "can we verify DeepSeek V4 at all?" to a more useful decision:
- When is DeepSeek V4 Flash the better fit?
- When is DeepSeek V4 Pro worth paying for?
- When should teams still choose GPT-5.4 or Claude Opus 4.6 instead?
If you want current route details and implementation guidance, start with the DeepSeek V4 API page.
TL;DR
- DeepSeek V4 Flash is now the cheapest officially documented option in this comparison set at $0.14 input / $0.28 output per 1M tokens, with 1M context and 384K max output. It is the strongest candidate for high-volume coding, agent routing, and cost-sensitive long-context workloads. DeepSeek Models & Pricing
- DeepSeek V4 Pro 0813 is the premium V4 route at $0.435 cache-miss input / $0.87 output per 1M tokens until August 16, 2026 16:00 UTC, when DeepSeek's published peak/off-peak repricing takes effect (off-peak $0.66 / $1.98, peak 2×). Independent aggregate results put it only slightly ahead of Flash 0731, so Pro should be evaluated as an escalation route rather than assumed to win every task. DeepSeek Models & Pricing Artificial Analysis
- GPT-5.4 remains the clearest officially documented OpenAI option for complex professional work, with 1,050,000 context, 128,000 max output, and $2.50 / $15.00 pricing. OpenAI Pricing OpenAI GPT-5.4 Model
- Claude Opus 4.6 remains a top-tier choice for coding and agentic tasks, with pricing at $5 / $25 per 1M tokens, 128K output, and 1M context in beta on the Claude Developer Platform. Anthropic Claude Opus 4.6
What is officially verifiable now
The table below uses only currently documented official vendor information.
| Topic | DeepSeek V4 Flash | DeepSeek V4 Pro | GPT-5.4 | Claude Opus 4.6 |
|---|---|---|---|---|
| Provider | DeepSeek | DeepSeek | OpenAI | Anthropic |
| Current documented version | DeepSeek-V4-Flash-0731 | DeepSeek-V4-Pro-0813 | GPT-5.4 | Claude Opus 4.6 |
| Input pricing | $0.14 / 1M cache miss | $0.435 / 1M cache miss | $2.50 / 1M | $5.00 / 1M |
| Cached input pricing | $0.0028 / 1M | $0.003625 / 1M | $0.25 / 1M | Pricing varies by caching and long-context tiers |
| Output pricing | $0.28 / 1M | $0.87 / 1M | $15.00 / 1M | $25.00 / 1M |
| Context window | 1M | 1M | 1,050,000 | 1M in beta |
| Max output | 384K | 384K | 128K | 128K |
| Thinking mode | Supported | Supported | Supported via reasoning effort | Supported via adaptive / extended thinking |
| Tool calls | Supported | Supported | Supported | Supported |
| Practical status | Best low-cost V4 route | Best higher-intelligence V4 route | Official OpenAI flagship route | Official Anthropic flagship route |
Sources: DeepSeek API Docs DeepSeek Models & Pricing OpenAI Pricing OpenAI GPT-5.4 Model Anthropic Claude Opus 4.6
Pricing reality check
The pricing story is now straightforward because all three vendors publish usable official numbers.
| Model | Input | Cached input | Output | Practical pricing takeaway |
|---|---|---|---|---|
| DeepSeek V4 Flash | $0.14 | $0.0028 | $0.28 | Cheapest official route here by a wide margin |
| DeepSeek V4 Pro | $0.435 | $0.003625 | $0.87 | Premium DeepSeek route at about 3.1× Flash's cache-miss input and output price |
| GPT-5.4 | $2.50 | $0.25 | $15.00 | Premium OpenAI route for complex professional work |
| Claude Opus 4.6 | $5.00 | context-tier dependent | $25.00 | Highest-cost route here, but still a top coding and agent model |
Two things stand out immediately:
- DeepSeek V4 Flash is the budget winner by a large margin.
- DeepSeek V4 Pro has moved from "unverified watchlist" to a real premium-but-still-cost-efficient option.
If cost per output token matters to your workload, the gap is especially important:
- DeepSeek V4 Flash output is far cheaper than GPT-5.4 and Claude Opus 4.6
- DeepSeek V4 Pro output is still well below GPT-5.4 and Claude Opus 4.6
The biggest practical change: DeepSeek V4 is now Flash vs Pro
Earlier DeepSeek V4 writeups treated V4 like one hypothetical model. That is no longer accurate for decision-making.
Today the more useful framing is:
- Choose Flash when cost, throughput, and broad deployment matter most
- Choose Pro when you want stronger reasoning quality but still want to stay below closed-model pricing
That makes DeepSeek V4 less like a single competitor to GPT-5.4 or Opus 4.6 and more like a two-tier product family.
Which model fits which workflow
Choose DeepSeek V4 Flash if you want the best cost-performance tradeoff
Flash is the strongest fit when you need:
- high-volume coding assistance
- cost-sensitive agent pipelines
- large-context document or repository ingestion
- routing defaults that must stay cheap
Because Flash keeps 1M context and 384K output while staying extremely inexpensive, it is now the easiest model in this comparison set to justify as a broad default route for many production systems. DeepSeek Models & Pricing
Choose DeepSeek V4 Pro if you want a premium V4 route without closed-model pricing
Pro is the stronger fit when you need:
- deeper reasoning than a budget route
- more difficult coding and analysis tasks
- longer-form structured output
- a step up from Flash without jumping all the way to Claude Opus 4.6 pricing
For many teams, Pro is not a universal replacement for GPT-5.4 or Opus 4.6. It is a lower-cost premium option worth evaluating side by side.
Choose GPT-5.4 if you want OpenAI's officially documented flagship route
GPT-5.4 remains attractive when you want:
- official OpenAI platform support
- a documented 1,050,000 context window
- 128,000 max output
- a familiar OpenAI developer workflow
The main tradeoff is still output pricing. GPT-5.4 is much more expensive than Flash and materially more expensive than Pro on output-heavy tasks. OpenAI Pricing
Choose Claude Opus 4.6 if your top priority is frontier Anthropic coding and agent work
Claude Opus 4.6 remains strong when you want:
- Anthropic's flagship coding model
- extended and adaptive thinking controls
- long-running agentic workflows
- Claude Platform features such as context compaction and 1M context beta
The main tradeoff is still cost. It is the most expensive route in this group. Anthropic Claude Opus 4.6
Context and output limits matter more than before
The DeepSeek V4 update also changes the long-context conversation.
Previously, one practical reason to choose GPT-5.4 or Claude Opus 4.6 was that DeepSeek V4 was not publicly documented. Now DeepSeek is officially in the same planning conversation because both Flash and Pro expose:
- 1M context
- 384K max output
That matters for:
- repo-scale code understanding
- long legal or research documents
- long multi-step agent loops
- tasks that need much larger single-response outputs
On pure max-output headroom, DeepSeek V4 now clearly beats GPT-5.4 and Claude Opus 4.6 based on current official docs.
Recommended decision by use case
| Use case | Best fit | Why |
|---|---|---|
| Need the lowest-cost official long-context route | DeepSeek V4 Flash | Cheapest official pricing here with 1M context and 384K output |
| Need a stronger premium DeepSeek route | DeepSeek V4 Pro | Higher-end V4 option without GPT / Claude-level pricing |
| Need an official OpenAI flagship model | GPT-5.4 | OpenAI-documented flagship with 1,050,000 context and 128K output |
| Need Anthropic's top coding and agent model | Claude Opus 4.6 | Strong Anthropic flagship with 1M beta context and 128K output |
| Need one model to evaluate for broad production routing | Start with Flash, then test Pro | Flash covers cost-sensitive routing; Pro covers harder workloads |
What changed from the old March 2026 conclusion
The old conclusion for this topic was:
- DeepSeek V4 was still a watchlist item
- V3.2 was the practical official DeepSeek baseline
- teams should not model budgets around V4 yet
That is no longer the correct conclusion.
As of August 13, 2026, the updated conclusion is:
- DeepSeek V4 Flash 0731 and Pro 0813 are the current documented API versions
- Flash and Pro should be evaluated directly on successful-task cost
- V3.2 is no longer the right planning baseline for V4-focused comparisons
FAQ
1. Is DeepSeek V4 officially available now?
Yes. DeepSeek's current pricing page lists
deepseek-v4-flash and deepseek-v4-pro, mapped to the 0731 and 0813 builds respectively. DeepSeek Models & Pricing2. Can I now compare DeepSeek V4 pricing with GPT-5.4 and Claude Opus 4.6 responsibly?
Yes. That is the main reason this article has been updated. DeepSeek now publishes official V4 pricing for both Flash and Pro, so the comparison no longer depends on rumors. DeepSeek Models & Pricing
3. Which DeepSeek V4 variant is the better first test: Flash or Pro?
For most teams, Flash is the better first test because it is dramatically cheaper while keeping 1M context and 384K max output. If your workloads are harder reasoning or coding tasks, then test Pro next.
4. Does GPT-5.4 still have an advantage?
Yes. GPT-5.4 still offers the official OpenAI flagship route, a documented 1,050,000 context window, 128,000 max output, and the surrounding OpenAI platform ecosystem. OpenAI GPT-5.4 Model
5. Does Claude Opus 4.6 still have an advantage?
Yes. Claude Opus 4.6 remains a top Anthropic choice for coding and agentic work, and Anthropic documents 1M context in beta plus 128K output. Anthropic Claude Opus 4.6
6. What is the cheapest officially documented option in this comparison?
DeepSeek V4 Flash is the cheapest officially documented model in this comparison set based on current official pricing. DeepSeek Models & Pricing
7. Should I still use DeepSeek-V3.2 as my budget baseline?
Not for V4 planning. If your topic is specifically DeepSeek V4, the better baseline is now V4 Flash and V4 Pro, because those are the models officially documented today.
8. Where should I go if I want route details and implementation guidance?
Use the DeepSeek V4 API page. This comparison article is for model selection, while the product page is the better place for route-level implementation details.
Sources
- DeepSeek API Docs
- DeepSeek Models & Pricing
- OpenAI API Pricing
- OpenAI GPT-5.4 Model
- Anthropic Claude Opus 4.6
- Anthropic Claude Opus 4.7
- Artificial Analysis: DeepSeek V4 Pro 0813
- Artificial Analysis: DeepSeek V4 Flash 0731
Ready to Evaluate DeepSeek V4?
Use the DeepSeek V4 API page to review current route details, pricing, and integration guidance for Flash and Pro.
