DeepSeek · tested in Taskade Genesis
DeepSeek V4 Pro in Taskade
DeepSeek V4 Pro from DeepSeek runs on every paid plan. We last tested it on Oct 8, 2026 in TSK-1, our app-building test.
- Status
- Available
- Plan
- Every paid plan
- Where to pick it
- More models or All Models
- Your own key
- No
- Last test
- Context window
- 1Mtokens
- Max output
- 384Ktokens
- Released
- Knowledge cutoff
- Where it runs
- 8services
TSK-1 scorecard for the DeepSeek family
TSK Score
DeepSeek family, across real apps made in Taskade Genesis
- Interfaceready to share
- Leading
- Taskcompletes your task
- Strong
- Memorykeeps your data
- Strong
- Adapthandles follow-up edits
- Emerging
The workspace champion with a delivery problem.
MoreLess
Pro still builds deep data. In September it put two apps behind a sign-in screen and left a third without a page. In October it built a family intake form that saved every answer and both photos, with no sign-in.
Published scores
Benchmark scores that DeepSeek and test sites published for DeepSeek V4 Pro. Taskade did not re-run them. Lab-reported marks a score from DeepSeek itself.
| Benchmark | Score | Source |
|---|---|---|
| SWE-Bench VerifiedresolvedHugging FaceLab-reported | 80.6% | Hugging FaceLab-reported |
| SWE-Bench Proresolved · preview checkpoint; max effortHugging FaceLab-reported | 55.4% | Hugging FaceLab-reported |
| Terminal-Benchaccuracy · v2.0 · preview checkpoint; max effortHugging FaceLab-reported | 67.9% | Hugging FaceLab-reported |
| LiveCodeBenchpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 93.5% | Hugging FaceLab-reported |
| BrowseComppass@1 · preview checkpoint; max effortHugging FaceLab-reported | 83.4% | Hugging FaceLab-reported |
| MCP Atlaspass@1 · preview checkpoint; max effort · publicHugging FaceLab-reported | 73.6% | Hugging FaceLab-reported |
| Toolathlonpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 51.8% | Hugging FaceLab-reported |
| GPQA Diamondpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 90.1% | Hugging FaceLab-reported |
More published scores (18)
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Coding Agent Indexaverage pass@1 · high · Claude CodeArtificial Analysis | 50.1% | Artificial Analysis |
| SWE-Atlas Codebase QnApass@1 · high · Claude CodeArtificial Analysis | 67.8% | Artificial Analysis |
| SWE-Bench Propass@1 · high · Claude Code · hard-aaArtificial Analysis | 18% | Artificial Analysis |
| Terminal-Benchpass@1 · v2.1 · high · Claude CodeArtificial Analysis | 64.7% | Artificial Analysis |
| MMLU-ProEM · preview checkpoint; max effortHugging FaceLab-reported | 87.5 | Hugging FaceLab-reported |
| SimpleQA-Verifiedpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 57.9% | Hugging FaceLab-reported |
| Chinese SimpleQApass@1 · preview checkpoint; max effortHugging FaceLab-reported | 84.4% | Hugging FaceLab-reported |
| Humanity's Last Exampass@1 · preview checkpoint; max effort; without toolsHugging FaceLab-reported | 37.7% | Hugging FaceLab-reported |
| Codeforcesrating · preview checkpoint; max effortHugging FaceLab-reported | 3,206 | Hugging FaceLab-reported |
| HMMTpass@1 · preview checkpoint; max effort · February 2026Hugging FaceLab-reported | 95.2% | Hugging FaceLab-reported |
| IMOAnswerBenchpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 89.8% | Hugging FaceLab-reported |
| MathArena Apexpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 38.3% | Hugging FaceLab-reported |
| MathArena Apex Shortlistpass@1 · preview checkpoint; max effortHugging FaceLab-reported | 90.2% | Hugging FaceLab-reported |
| MRCRMMR · preview checkpoint; max effort · 1M contextHugging FaceLab-reported | 83.5 | Hugging FaceLab-reported |
| CorpusQAaccuracy · preview checkpoint; max effort · 1M contextHugging FaceLab-reported | 62% | Hugging FaceLab-reported |
| SWE-Bench Multilingualresolved · preview checkpoint; max effortHugging FaceLab-reported | 76.2% | Hugging FaceLab-reported |
| Humanity's Last Exampass@1 · preview checkpoint; max effort; with toolsHugging FaceLab-reported | 48.2% | Hugging FaceLab-reported |
| GDPval-AAElo · preview checkpoint; max effortHugging FaceLab-reported | 1,554 | Hugging FaceLab-reported |
LMArena scores
| Benchmark | Score | Source |
|---|---|---|
| Text arena±4 · 54,902 votes · Rank 71 of 416 · LMArena | 1,455 | LMArena |
| WebDev arena±6 · 13,157 votes · Rank 69 of 145 · LMArena | 1,465 | LMArena |
Scores from the LMArena leaderboard dataset (CC BY 4.0). People vote between two answers, and the votes set the score. Taskade picked the Overall category and the default setting of each model.
Scores collected by models.dev (MIT). Each source links the page that published the score.
Versions tested
Open a version to read its findings.
DeepSeek V4 ProTested Oct 202611 findings
The workspace champion with a delivery problem. Pro still builds deep data. In September it put two apps behind a sign-in screen and left a third without a page. In October it built a family intake form that saved every answer and both photos, with no sign-in.
The richest workspace of the test: 8 automations, a record 60 fields, and all four builds word for word.
Won the tracker test as the most efficient app that opened. Two stronger-looking results could not be used, so they did not receive a score.
Went from none of its four builds matching the brief to all four word for word on a shorter prompt. The biggest jump we have measured in a single test.
The cheapest complete match tracker of the test.
Built the deepest lead table of the sales CRM test, with 12 fields per lead, then put its one-page board behind a sign-in screen.
Finished the client sign-up form with all 32 questions word for word and left both automations switched off.
Spent the fleet-inspection test building data and automations and wrote no page.
Built a complete match-tracker data layer, then put the app behind a sign-in screen.
Built one of the two best-looking match trackers of its test, with a win-rate trend chart in light and dark, in 36 minutes, but its hero picker listed only 17 heroes.
Built the family intake form in 7 minutes 24 seconds so it saved every answer and both photos, and the new teacher question landed. Two sample families showed as real, and its summary did not mention them.
Built a Dota 2 match tracker with a win-rate trend that saved a logged match and added the heroes page when asked. About 15 sample matches showed as real.
DeepSeek V4 Pro, max thinkingTested Oct 20263 findings
DeepSeek V4 Pro with its deepest thinking setting, which the Taskade model picker no longer offers. The 15-minute test limit ended its build before a closing summary, and the follow-up turn finished an intake form that saved every answer and both photos with no fake sample families.
The 15-minute test limit ended the family intake build before a closing summary. The follow-up turn finished the app and added the new teacher question.
The finished form saved every answer and both photos, and the teacher page started empty instead of showing made-up families.
Built the best-looking Dota 2 match tracker of its set, and a logged match saved. The 15-minute limit ended the build, and the follow-up turn finished the interface and the heroes page.
DeepSeek V4 Pro, high thinkingTested Oct 20262 findings
DeepSeek V4 Pro with a high thinking setting, which the Taskade model picker no longer offers. Its family intake form could never keep a photo, so no parent could submit it. Its match tracker labeled its demo matches and kept them out of the stats.
The 15-minute test limit ended the family intake build before a closing summary, and the form dropped every photo a parent picked, so the two-photo rule blocked every submission.
Built a Dota 2 match tracker that saved a logged match, labeled its demo matches and kept them out of the stats, and added the heroes page when asked.
Results by task
DeepSeek V4 Pro
Model facts
The token limits are the most the model accepts. Taskade can run it with lower limits.
- Last updated
- Input types
- Text
- Output types
- Text
- Reasoning
- Yes
- Tool calling
- Yes
- File attachments
- No
- Structured output
- Yes
- Open weights
- Yes
- Reasoning effort
- NoneLowHighMax
Published specs from the model maker, via models.dev (MIT).
Where it runs
Services outside Taskade that serve DeepSeek V4 Pro, from models.dev. Each link opens in a new tab.
How to choose DeepSeek V4 Pro
- In the builder chat, open the model menu and choose More models. Then pick the model.
- In the agent editor, pick the model under All Models.
- If your model menu does not show the model, Taskade uses TSK-Auto instead.
Where DeepSeek ranks
TSK Score out of 100.
Compare DeepSeek
Other DeepSeek models
FAQ
Which Taskade plans include DeepSeek V4 Pro?
DeepSeek V4 Pro runs in Taskade on every paid plan. If your model menu does not show it, Taskade uses TSK-Auto instead.
Has Taskade tested DeepSeek V4 Pro?
Yes. We last tested DeepSeek V4 Pro on Oct 8, 2026 in the TSK-1 Benchmark. The workspace champion with a delivery problem. Pro still builds deep data. In September it put two apps behind a sign-in screen and left a third without a page. In October it built a family intake form that saved every answer and both photos, with no sign-in.
How do I choose DeepSeek V4 Pro in Taskade?
In the builder chat, open the model menu and choose More models. Then pick DeepSeek V4 Pro. In the agent editor, pick it under All Models.
Who makes DeepSeek V4 Pro?
DeepSeek makes DeepSeek V4 Pro. Taskade runs it on every paid plan, beside frontier models from other top AI labs.