iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
For new API projects, start with deepseek-flash, the current API name for DeepSeek-V4.1-Flash-1017. DeepSeek’s official material does not provide a controlled, same-method comparison against V3, so it cannot support a definitive “1017 is better” or “V3 is cheaper” claim. Use your own representative prompts, measure latency and output quality, and verify the live bill before committing. Treat V3 as a legacy option whose current availability and pricing must be confirmed with DeepSeek.
The short decision
The practical choice depends first on whether you are starting a new integration or maintaining an existing one:
- New API integration: use the model identifier
deepseek-flash. DeepSeek’s API pricing table identifies that alias as DeepSeek-V4.1-Flash-1017, and the October 17, 2026 changelog says the latest version continues to be called with that value. - Existing V3 application: do not switch solely because a newer name exists. Capture a test set from your current workload, run it through
deepseek-flash, and compare correctness, latency, output length, tool behavior and the actual invoice. - Considering V3 for a new deployment: first confirm that DeepSeek will provision V3 for your account and route, then obtain its current rates. The current public pricing table reviewed for this article lists
deepseek-flashanddeepseek-v4-pro, not V3; that omission does not prove that V3 is unavailable everywhere. - Long-context or tool-using work: 1017 has documented support for a 1-million-token context, up to 384,000 output tokens, thinking and non-thinking modes, JSON output, tool calls, the Responses API and the Anthropic API. Do not assume that V3 has the same limits without current V3 documentation.
In other words, 1017 is the sensible default for a fresh project because it is the currently documented model behind the active alias. V3 can still be the right answer for a proven, compatible workload, but only after an availability, price and regression check.
Recommended Free Tools
What “V4.1-Flash-1017” and “V3” mean in the API
The current 1017 alias
DeepSeek’s October 17, 2026 changelog says V4.1 Flash was upgraded to DeepSeek-V4.1-Flash-1017. The official web and app models were updated at the same time. The API calling method and price did not change, and the changelog instructs users to continue calling the latest model with deepseek-flash. The API pricing table also identifies deepseek-flash as version DeepSeek-V4.1-Flash-1017.
Popular on Amazon
- Powerful Turbo Fan:CAGIWIRU Electric Air Duster have reached speed up to 130000RPM,Can efficient remove dust and debris.With LED light and Three Adjustable Speed to meet with different cleaning tasks.
- With Attached USB C Fasting Charging cable.Charing 3 hours,Enjoy up to 240 minutes of use on the lowest setting, For third speed can use for 25mins.with four charging options to meet with your needs.
- Wide Application: attached 5 different nozzles,making it suitable for a variety of scenes, such as car,pet, pc, keyboards, and other electronic devices. It also serves for office and home clean duster.
- Handy Design & Portable:Cordless air duster electric nozzle is Light weight and small size, design and comfortable to hold and space saving, convenient to carry around. You can put it in your room or inside the car without taking up space.
- Economical &Reusable: Built durable materials with a long-lasting battery,our CAGIWIRU air duster it's a sustainable alternative to disposable air cans.
This distinction matters operationally: your application normally sends the alias, not the long release label. An alias that moves to a newer release can change outputs even when your code is unchanged. Record the date and model information for every evaluation, watch the changelog, and rerun regression tests after a vendor upgrade.
V3 is a launch-era model description, not a current service contract
DeepSeek’s December 26, 2024 V3 announcement described a 671-billion-parameter mixture-of-experts model with 37 billion parameters activated per token, trained on 14.8 trillion tokens. Those figures describe the launch model and training run. They do not establish today’s API context window, output cap, routing, latency or support status for a V3 endpoint.
Documented capabilities of DeepSeek-V4.1-Flash-1017
The current API table documents the following properties for deepseek-flash at the time covered by the October 2026 documentation. Capability and limit pages can change, so check the live table when you implement or renew a deployment.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →| Property | What the official documentation states | What it means for a decision |
|---|---|---|
| API model value | deepseek-flash |
Use this value when calling the current 1017 model. |
| Documented version | DeepSeek-V4.1-Flash-1017 | The alias is the stable integration name; the release label identifies the version behind it. |
| Context length | 1M tokens | Useful for large repositories, long transcripts and multi-document prompts, subject to your real token usage and billing. |
| Maximum output | 384K tokens | A ceiling in the API table, not a promise that every request will produce that much text or that doing so is economical. |
| Reasoning modes | Thinking and non-thinking | Measure both modes on your workload; they can trade response time, cost and depth. |
| Structured output | JSON output | Suitable for extraction and machine-readable pipelines when your prompt and validation handle malformed or incomplete responses. |
| Tool use | Tool calls | Relevant to agents and function-driven applications; test argument accuracy and recovery from failed tools. |
| API interfaces | Responses API and Anthropic API | May reduce adapter work when an application already targets one of those interfaces. |
| Beta interfaces | Chat Prefix Completion; non-thinking-only beta FIM Completion | Beta behavior should be isolated behind an adapter and tested for compatibility before production use. |
DeepSeek’s October update also recommends DeepSeek Harness 0.2.1-rc.1 or later for the best experience with the new version, and names DeepSeek Harness, Tencent CodeBuddy and WorkBuddy as optimized platforms. This is vendor guidance, not an independent compatibility certification. If your workflow uses one of these tools, test the exact release and configuration you plan to operate.
What the published 1017 benchmarks do—and do not—show
The October 17 update reports that V4.1-Flash-1017 is stronger on the listed benchmarks than the prior V4.1 Flash version, with faster completion and shorter outputs. The figures below are vendor-published results. Each benchmark has its own task, harness, prompting and scoring method; scores from different rows are not interchangeable.
| Benchmark | Reported 1017 result | Qualification in the source |
|---|---|---|
| HLE without tools | 42.6 | DeepSeek API documentation, 2026. |
| HLE, starred pure-text subset | 44.0 | The starred value is specifically for the pure-text subset. |
| Terminal-Bench 4.0 | 42.3 | Run with the Claude Code framework. |
| DeepSWE v1.1 | 77.0 | Run with Mini-SWE-V2. |
| ProgramBench | 29.1 | Vendor-reported benchmark result. |
| NL2Repo-Bench | 68.2 | Vendor-reported benchmark result. |
| Automation-Bench v1.0.6 | 60.5 | Vendor-reported benchmark result. |
| Agents’ Last Exam | 35.4 | Vendor-reported benchmark result. |
| Apex Bench v1.1 | 72.1 | Vendor-reported benchmark result. |
| Chartography with tools | 83.0 | Tool-enabled result. |
| BabyVision with tools | 92.5 | Tool-enabled result. |
| ZeroBench-main with tools | 51.0 | Uses Pass@5. |
None of these numbers is a V3 baseline produced under the same conditions. They therefore cannot answer the literal head-to-head question. A 77.0 on DeepSWE, for example, says nothing by itself about how much better or worse 1017 is than V3 on your repository, because the V3 result, prompt setup and tool framework are not supplied.
Also popular on Amazon
- Screen Cleaner is Your Screen’s New Bestfriend! - EVEO is proud to present a screen cleaner that’s perfectly suited for your TV. Meet your new screen cleaner, computer screen cleaner, laptop cleaner, iPad cleaner, Macbook,tv cleaner or any other electronic device. Our screen cleaner spray and wipe along with the included plush microfiber cloth, easily removes dust, fingerprints & other smudges on the screen’s surface. Clean effectively for a streak-free screen
- Gentle on your Screens - Our screen cleaner spray and wipes is compatible to be used even for the most sensitive LCD, LED, CRT, and OLED screens in proper use. Be at ease knowing that this product is compatible with all the major brands: LG, SONY, Samsung, Sharp, iPad, iPhone, Android screens, Kindle, PC monitors - we’ve got you covered! Can be used as LCD screen cleaner, tv cleaner, smart tv screen cleaner, and more
- Premium Super Soft Plush Microfiber Cloth - Included in this efficient screen cleaning kit. This screen cleaner spray and wipe has our signature plush microfiber cloth specially designed to comply with screens and monitors and leave them streak-free. Has been shown to effectively absorbs stubborn stains, zero streaks, and no fiber shedding. Always use this together with our screen cleaner for the best results
- Easy-to-Use Screen Cleaner Kit + microfiber cleaning cloth - The EVEO Screen Cleaning Kit will easily be your new favorite tool when it comes to taking care of the things you love most, quick and easy solution for a streak-free screen. Simply screw on the sprayer gun on the screen cleaner bottle. Spray the cleaning liquid onto the microfiber cloth, then wipe the screen until desired cleanliness & shine is achieved.
- Quality Screen Care Kit Worth Purchasing - Enjoy a convenient Screen Cleaner anytime, anywhere you need it. Your satisfaction is our top priority. We stand behind the quality of our products and that’s why, you shouldn’t be worrying at all. We manufacture the Best Screen Cleaner. and can be used for laptop cleaning kit, laptop screen cleaner, tv screen cleaner for smart tv. MacBook screen cleaner, monitor cleaner or screen cleaner spray for electronic devices with microfiber cleaning cloth
What is actually known about V3
Architecture and training figures
The V3 launch announcement reported 671B total parameters, 37B active parameters per token and 14.8T training tokens. These are useful historical context when reading older technical coverage, but they are not substitutes for current API documentation. Parameter counts do not predict your application’s factual accuracy, latency or bill.
Historical launch pricing
The same announcement listed rates from February 8 onward of $0.27 per million input tokens for cache misses, $0.07 per million input tokens for cache hits and $1.10 per million output tokens. Those were launch-era V3 rates. They must not be presented as current prices or used to calculate a present-day budget without confirmation from DeepSeek.
Current availability is unresolved
The current pricing page reviewed for this comparison lists deepseek-flash and deepseek-v4-pro, but not V3. That observation is not proof that V3 is unavailable by every route or for every customer. Before choosing V3, ask DeepSeek or check the account-specific model list for three separate facts: whether the model can be called, which endpoint or interface serves it, and which current input, cached-input and output rates apply.
Why an official winner has not been established
A fair model comparison requires the same task set, prompt policy, tool definitions, sampling settings, context size, hardware or service route, scoring rubric and date. The material available here compares different releases and supplies no controlled V3-versus-1017 run. V3’s figures come from a December 2024 launch announcement, while 1017’s benchmark claims are from October 2026 documentation and use named frameworks on some tasks.
That leaves several ways a seemingly simple comparison can mislead:
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute- A newer benchmark may measure a different skill, so a higher score is not evidence of a universal improvement.
- Tool-enabled and no-tool results are different experiments.
- Pass@5 is not directly comparable with a single-attempt score.
- Thinking and non-thinking modes can produce different quality and latency profiles.
- A model that wins a public coding benchmark may be slower or less reliable on your codebase, schema or language mix.
- Historical V3 rates may no longer match the route or account that would serve V3 today.
Compare the models on the dimensions that affect your system
1. Availability and routing
For 1017, the documented route is the deepseek-flash model value. For V3, establish access before doing extensive engineering. A model that cannot be provisioned, or that is available only through a different interface, is not a practical production alternative no matter how attractive its historical specifications look.
Also popular on Amazon
- 【Large size】：Measuring 800 x 300 mm/ 31. 5×12 inches, which is wider and taller than others. Large mouse pad can perfectly fit your keyboard and mouse of any size, and there is also ample space for peripherals such as phones, tablets, etc.
- 【LED backlight mouse pad】：Blade Hawks RGB mouse pad provides 7 static modes and 7 dynamic modes will give you the options you are looking for. Power-off memory function remembers the last lighting mode you selected. The super glow fiber Chroma will give you a colorful gaming setup that you want.
- 【Anti-slip rubber base】：Made from Natural rubber, this gaming mousepad will firmly grip your desktop, allowing you to enjoy the ultimate precision in the games you are most passionate about. It also comes with a texture that enhances this ability even further.
- 【Smooth and waterproof surface】: Micro-textured cloth surface, for extremely precise mouse control and a smooth movement. When liquid splashes on the gaming mouse pad, it forms water droplets and slides down. Will not delay your work or gameplay.
- 【Plug and play】: This mouse and keyboard pad is powered by USB, plug and play, no driver required. One button control light modes. Press one time to change the lighting mode, press twice to change the brightness, press and hold about 3 seconds turn ON/OFF.
2. Task quality on your data
Build a fixed evaluation set from real requests: successful examples, known failures, long-context cases, structured outputs and adversarial inputs. Score factual correctness, required fields, compilation or test results, citation behavior and refusal or escalation handling. Keep prompts and tool schemas identical, and label every run with the model, mode and date.
3. Latency and output length
Measure time to first token and total completion time separately. Also record output tokens. DeepSeek says 1017 completes faster and produces shorter outputs than the prior V4.1 Flash version, but that is not a V3 comparison and may not predict your network, prompt or tool workload.
4. Cost under your traffic pattern
Token prices alone are insufficient. Count uncached input, cached input and output tokens for each request, then multiply by the current rates for the exact model and route. Long prompts, retries, tool messages and generated JSON can dominate a bill even when the per-million-token rate appears low. Use the live pricing table rather than the historical V3 numbers quoted above.
5. Context and interface requirements
If you need a million-token context, 384K maximum output, Responses API compatibility, Anthropic API compatibility, JSON output or tools, 1017 has those properties documented. V3’s equivalent limits were not established in the available current material. Require written, current documentation or a successful account test before designing around a V3 capability.
6. Operational fit
Run the model in the actual harness or coding agent, not just a bare chat test. DeepSeek’s recommendation of Harness 0.2.1-rc.1 or later and its references to Tencent CodeBuddy and WorkBuddy indicate intended integrations, but they remain vendor statements. Check tool-call schemas, streaming behavior, error handling, rate limits and observability in your chosen platform.
A repeatable V3-versus-1017 evaluation
The following process gives you evidence without pretending that public benchmark scores settle the choice.
Also popular on Amazon
- 【Softer and More Comfortable】Vaydeer wrist rest has unique diamond pattern, which is the combination of softness and aesthetics. The materials of wrist rest are improved into higher quality memory foam and covered with silky smooth lycra. The computer wrist rest makes you as comfortable and cushiony as like rest your wrists on clouds.
- 【Ergonomic Wrist Saver】The wrist rests for keyboard and mouse comes with a 17.32×3.15×0.83 inch keyboard wrist pad and a 5.94×3.15×0.83 inch mouse wrist support. Based on ergonomic design, the unique concave shape is the perfect fit for your wrist joints. The wrist rest pad fits most computer keyboards and laptops, improve hand and wrist posture, release your wrist and arm stress.
- 【Non-Slip Rubber Bottom】Featuring an anti-skid silicone base on the bottom, this wrist keyboard support stays firmly in place on your desk, preventing the padding from sliding around, ensuring stable and consistent wrist support during extended computer sessions.
- 【Better Experience & Pain Relief】Our keyboard arm rest is beneficial to alleviate the soreness caused by direct contact and friction between your arm and a hard desk surface, reducing the risk of wrist fatigue or carpal tunnel. The soft texture of memory foam can evenly distribute the pressure around your wrists and provide good support with just enough give.
- 【Helpful in Multiple Scenarios】Whether you're working, studying, writing, typing, gaming, this keyboard and mouse rest combo is an essential accessory to add comfort and support to your hands and wrists. It’s also a great gift for men, women, family, friend, coworker, gamer, teacher, etc.
- Freeze the workload. Export a representative, privacy-approved sample. Include easy, typical and failure-prone cases; separate interactive requests from batch jobs and agent loops.
- Define success before running. Decide what counts as correct, how to score partial answers, which JSON fields are mandatory, what test suite must pass and what latency or cost ceiling is acceptable.
- Record access details. For 1017, record
deepseek-flash, the API interface, thinking mode, date and any harness version. For V3, record the exact model identifier and route supplied by DeepSeek; do not infer them from the 2024 announcement. - Keep the request shape constant. Use the same system instructions, user prompts, context, tool definitions, temperature or equivalent sampling controls and retry policy. Change only the model and the mode being tested.
- Validate structured responses. Parse JSON, check required keys and reject invalid values. A fluent response that fails your parser is a production failure.
- Measure quality and reliability. Have a rubric or executable test score each answer. Track pass rate, hallucinated facts, tool-call argument errors, empty responses, retries and timeouts.
- Measure performance. Capture time to first token, completion time, input tokens, cached-input tokens, output tokens and total request count. Keep warm-up and network effects consistent.
- Calculate the real bill. Apply the current rates for the exact route to the recorded token counts. Include retries and tool turns. Do not substitute the historical V3 launch rates for a current quote.
- Test long context separately. Increase context in controlled steps and note retrieval accuracy, latency, truncation and cost. A nominal 1M-token limit does not guarantee useful reasoning across a million tokens.
- Make a staged decision. Send a small percentage of low-risk traffic to the winner, monitor regressions and retain a rollback path. Re-run the suite when the
deepseek-flashalias changes version.
Minimal request record
Keep a machine-readable record alongside each result. The essential fields are:
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match{
"model": "deepseek-flash",
"model_version_label": "DeepSeek-V4.1-Flash-1017",
"mode": "thinking or non-thinking",
"prompt_set": "fixed evaluation identifier",
"input_tokens": 0,
"cached_input_tokens": 0,
"output_tokens": 0,
"time_to_first_token_ms": 0,
"completion_time_ms": 0,
"quality_score": 0
}Replace the example values with measurements from your system. For a V3 run, preserve the exact identifier and version information returned by the service instead of copying the 1017 labels.
Cost and billing questions to settle before production
Do not use the old V3 rates as a forecast
The February 8 launch rates—$0.27 per million cache-miss input tokens, $0.07 per million cache-hit input tokens and $1.10 per million output tokens—are historical. They are useful for understanding what DeepSeek announced in 2024, not for approving a 2026 budget.
Ask for the complete current price matrix
For each candidate, obtain the current prices for uncached input, cached input and output, and ask whether rates vary by interface, region, account tier or service status. Confirm how cache hits are defined, whether tool messages count as input, how failed or retried calls are billed and whether beta interfaces have separate terms.
Model a realistic request
Suppose one request sends a large system prompt, a document, several tool results and a short question, then receives a long structured answer. The bill depends on every token sent and generated, not only the user’s final sentence. Build a spreadsheet or meter that records token classes per request and multiplies them by the current table. Recalculate after prompt, tool or retry changes.
Free tools Windows power users keep installed
One-click scans. No signup required.
Context, tools and coding workflows
Long documents and repositories
1017’s documented 1M-token context can accommodate workloads that exceed smaller windows, but capacity is not the same as retrieval quality. Test whether the model finds the right passage, respects instruction priority and remains consistent when irrelevant material fills the window. Chunking, indexing and selective retrieval may still reduce cost and latency.
Also popular on Amazon
- 【No Drill Mounted】Cable management tray can be assembled in as little as 3 minutes or less-simply. Side clips easily clampe the edge of desk, pass through the holes on the sides, and secure excess cables with baskets. Desk cable management supports inward or outward installation to meet your needs in different scenarios, and provides you with great convenience when collecting and organizing wires.
- 【Rubber Pad Protects Desk】Wire organizer under desk built-in soft rubber pads on the clamp protect your desk from scratches, so you will not damage your furniture or office.
- 【Sturdy and Beautiful】Cord management under desk is made of high-quality carbon steel and holds about 15 lbs. The wire basket has 2 holes on each side of the tray to enable your cables to pass freely. Cable baskets are suitable for desk thickness use from 0.4" to 2.4".
- 【Space Saving & Keep Organized】Desk cable management tray holds all your power cords, outlet strips, USB hubs and computer transformers hidden from your feet, and the data cables will not enter the vacuum cleaner to keep your desk clean and tidy. Desk wire organizer is also suitable for use in the kitchen and outdoors.
- 【Reduce Safety Risks】Cable tray under desk keeps all plugs away from your kids, no more worry about tripping. Using 2 holes wire tray saves extra space, helps you easily cabling and wire protection. The mesh design is not easy to accumulate dust, but avoids safety accidents caused by messy wires.
Agent and terminal tasks
The reported Terminal-Bench 4.0 score of 42.3 used the Claude Code framework, and the DeepSWE v1.1 score of 77.0 used Mini-SWE-V2. Those qualifiers are part of the result. If your agent uses different tools, permissions or test commands, reproduce your environment rather than treating the scores as expected completion rates.
Function and tool calls
Because the current API table documents tool calls, evaluate more than whether a tool is selected. Check argument schema validity, missing arguments, duplicate calls, recovery after a tool error, handling of partial results and whether the model stops when the task is complete. Compare the number of tool turns as well as the final answer, because extra turns affect both latency and cost.
Structured and code output
For JSON pipelines, validate against a schema and log repair attempts. For code, run compilation, unit tests, static analysis and security checks. A benchmark score or an attractive sample is not a substitute for executing the artifacts your application will ship.
Migration guidance for existing V3 users
- Inventory dependencies. List every place the V3 identifier, output format, context assumption, retry policy and token budget appears.
- Capture a baseline. Save representative V3 prompts and outputs, along with correctness scores, latency, token counts and current billing records.
- Adapt the integration to the documented alias. Test
deepseek-flashthrough the interface you intend to operate. Do not assume that a V3-specific parameter or limit exists on 1017. - Recheck prompts and parsers. A shorter or differently structured answer can expose assumptions in regular expressions, JSON parsing, stop conditions or downstream validators.
- Run shadow traffic. Send copies of approved requests to 1017 without changing user-visible results. Compare quality and cost before a live cutover.
- Roll out gradually. Start with low-risk traffic, monitor error classes and keep the V3 route only while its availability and billing are confirmed.
The October 17 changelog says the API calling method and price did not change for the 1017 upgrade, which reduces mechanical migration work. It does not guarantee identical outputs, latency or token counts, so behavioral testing remains necessary.
Common mistakes and how to avoid them
- Calling the public benchmark list a head-to-head test. It is not. The 1017 figures do not include a same-method V3 baseline.
- Quoting 2024 V3 prices as current. Label them historical and obtain a current quote.
- Assuming V3 is either gone or guaranteed. The current table’s omission is evidence to verify, not a universal availability statement.
- Comparing unlike modes. Keep thinking and non-thinking runs separate and record which mode produced each result.
- Ignoring tool and harness differences. Framework-specific scores may not transfer to your agent.
- Designing around undocumented V3 limits. Require current documentation or an account test for context, output and interface behavior.
- Testing only polished prompts. Include malformed input, long context, retries, empty tool results and ambiguous requests.
- Forgetting alias changes. Treat a model-alias upgrade as a regression-test event even when the request code remains unchanged.
Recommendation by situation
| Your situation | Recommended starting point | Reason and required check |
|---|---|---|
| Building a new API service | deepseek-flash (DeepSeek-V4.1-Flash-1017) |
It is the currently documented API model and has published limits and interfaces. Verify live prices and limits before launch. |
| Maintaining a V3 production system that works | Keep V3 temporarily while testing 1017 | Protect a known-good workload; compare on your own prompts and confirm whether V3 remains provisioned. |
| Choosing mainly on price | Neither until you have current quotes | The available V3 figures are historical, and no current 1017 price numbers are stated here. Calculate from your measured token mix. |
| Need a documented million-token context or 384K output ceiling | 1017, subject to workload testing | Those limits are documented for deepseek-flash; equivalent V3 limits were not established. |
| Running an agent or coding harness | 1017 with a compatibility pilot | DeepSeek names Harness 0.2.1-rc.1 or later and certain coding platforms as optimized, but you must test your exact tools and permissions. |
| Seeking a benchmark-proven V3 advantage | No conclusion from the published material | A controlled, current V3-versus-1017 comparison is not provided. |
Final answer
Choose deepseek-flash—DeepSeek-V4.1-Flash-1017—for a new integration, because it is the current documented alias with explicit 1M-context, 384K-output, reasoning, JSON, tool and API-interface support. Do not claim that it universally outperforms V3: the official figures available do not constitute a controlled comparison. If V3 is already embedded in your system, keep it long enough to run a matched evaluation and verify its current access and billing. The model that should remain in production is the one that meets your quality, latency, reliability and cost thresholds on your own traffic.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
