none reviewed yet · 4 benchmark observations · latest capture Sep 8, 2026
Inspect the evidenceDeepSeek · MODEL PROFILE
DeepSeek V4 Pro 0813
Cost-controlled agentic backend and full-stack work, and teams that need MIT-licensed weights for compliance
Saved editorial fit for your current task lens; not a benchmark result
How scoring worksVerified rates (Sep 8, 2026) · 100 runs / month at your token settings
Adjust workloadAt a glance
9 facts- Provider
- DeepSeek
- Type
- API model
- Access
- DeepSeek API (OpenAI- and Anthropic-compatible base URLs), Alibaba Cloud Model Studio, OpenRouter (18 providers incl. StreamLake, NovitaAI, Baidu Qianfan, DeepInfra, CoreWeave), self-hosting
- Context
- 1,000,000 tokens
- Maximum output
- 384K tokens (shared 384k output budget)
- License / weights
- Open weights
- Input modalities
- text
- Output modalities
- text
- Pricing notes
- DeepSeek direct peak: $1.32 in / $3.96 out per 1M; off-peak exactly half at $0.66/$1.98; cache-hit input $0.044 peak / $0.022 off-peak. Peak hours 01:00-04:00 and 06:00-10:00 UTC Mon-Fri. OpenRouter blended: $0.5795 in / $1.738 out, cache read $0.01932.
The practical trade-offs
Documented strengths
- Terminal-Bench 2.1 87.9% is within 0.9 pts of the best model in the set while costing a fraction of it
- MIT-licensed 1.6-1.7T MoE weights, so the exact model can be self-hosted
- Sub-2-second TTFT and 1M context on the hosted API
- Off-peak pricing halves an already low rate ($0.66/$1.98)
Limitations to check
- Design Arena Website 1258 (rank 42) shows visual taste well behind frontier models
- Text-only: screenshot-to-code needs the separate vision-exp model
- SWE-bench Pro 55.4% is far below the Anthropic tier
- Serving 1.6T+ weights locally is out of reach for most teams
Follow the source
9 source linksEvidence & conditions
9Source links
NoneReviewed records
4Benchmark runs
Sep 8, 2026Latest capture
- Evidence source 1
api-docs.deepseek.comDate not recorded - Evidence source 2
huggingface.coDate not recorded - Evidence source 3
openrouter.aiDate not recorded - Evidence source 4
arena.aiDate not recorded - Evidence source 5
benchlm.aiDate not recorded - Evidence source 6
benchlm.aiDate not recorded - Evidence source 7
benchlm.aiDate not recorded - Evidence source 8
artificialanalysis.aiDate not recorded - Evidence source 9
docs.qwencloud.comDate not recorded
Benchmark observations
Terminal-Bench 2.1 Inspect conditions
- name
- Terminal-Bench 2.1
- value
- 87.9
- unit
- %
- source Url
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
- checked At
- 2026-09-08T19:28:18Z
- attribution
- provider-reported
- configuration
- Exact DeepSeek-V4-Pro-0813; minimal DeepSeek Harness; reasoning_effort=max; temperature=1; top_p=0.95.
DeepSWE Inspect conditions
- name
- DeepSWE
- value
- 62.7
- unit
- %
- source Url
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
- checked At
- 2026-09-08T19:28:18Z
- attribution
- provider-reported
- configuration
- Exact DeepSeek-V4-Pro-0813; minimal DeepSeek Harness; reasoning_effort=max; temperature=1; top_p=0.95.
NL2Repo Inspect conditions
- name
- NL2Repo
- value
- 61.5
- unit
- %
- source Url
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
- checked At
- 2026-09-08T19:28:18Z
- attribution
- provider-reported
- configuration
- Exact DeepSeek-V4-Pro-0813; minimal DeepSeek Harness; reasoning_effort=max; temperature=1; top_p=0.95.
Arena WebDev Frontend Inspect conditions
- name
- Arena WebDev Frontend
- value
- 1586
- unit
- arena_score
- source Url
- https://arena.ai/leaderboard/code/webdev/frontend
- checked At
- 2026-09-08T19:28:18Z
- attribution
- benchmark-publisher
- configuration
- Exact row deepseek-v4-pro-high-20260813; source results dated 2026-09-05; human preference; confidence interval +/-12; 2718 votes. This is a checked source snapshot, not a newly run September 8 evaluation.
Source reports, measured results, and editorial assessments are separate evidence types. Source dates do not imply a new benchmark run. Capture dates are shown when recorded; otherwise we say the capture date was not recorded. Recorded dates are compared with this catalogue snapshot (Sep 11, 2026), not today’s date.
Method AIS-V2
How this page is assessed
Quality claims follow Method AIS-V2.
Money never buys rank — commercial relationships cannot change rank, eligibility, evidence quality, or publication timing.
Submit or claim a listing is free and does not skip the editorial queue.
Private · this device only