Public benchmarkExternal source reviewReviewed by DanPublished Source checked Updated
DeepSeek V4.1 Flash: public coding benchmark receipts
The report
Verdict
Published benchmark evidence
The 74.2% DeepSWE and 31.2% Terminal-Bench figures refer to original V4.1 Flash, not the later Flash-1017 revision; ProgramBench 20.3 is not labeled as a percentage.
On the tier list: The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
What the benchmarks report
- DeepSWE v1.1: 74.2%. Original V4.1 Flash, not Flash-1017; Mini-SWE-V2. Reported by DeepSeek vendor comparison table.
- Terminal-Bench 4.0: 31.2%. Original V4.1 Flash, not Flash-1017; Claude Code harness. Reported by DeepSeek vendor comparison table.
- ProgramBench: 20.3. Original V4.1 Flash; metric as published. Reported by DeepSeek vendor comparison table.
Original sources
Source 1 — DeepSWE v1.1, Terminal-Bench 4.0, ProgramBench.
What this evidence tells a reader
The cited model card belongs to Flash-1017 but contains a comparison column for original V4.1 Flash. This page deliberately draws 74.2% DeepSWE v1.1 and 31.2% Terminal-Bench 4.0 from the original column. Anyone using a service alias needs to verify which checkpoint the service actually routes to before using either figure.
The DeepSWE result names a Mini-SWE-V2 harness, while Terminal-Bench uses Claude Code. The score gap reflects different tasks and tool setups, not a controlled demonstration of a coding weakness. ProgramBench is reported as 20.3 with the publisher's scale; adding a percent sign would invent a metric.
DanManREAL has separate DeepSeek usage-cost and session-audit records. Those firsthand data answer billing and workflow questions, not the benchmark's success rate. They are useful complements when deciding whether a cheap model is economical after retries and human repair.
How to use these results
- Resolve the precise Flash or Flash-1017 checkpoint before comparing benchmark figures.
- Measure successful accepted tasks and end-to-end spend rather than only tokens and score headlines.
- Keep ProgramBench's published metric notation intact until its scoring scale is independently established.
Limits of this report
- The provider comparison table and the local deepseek-flash alias do not establish an identical checkpoint.
- No same-task price-versus-correctness comparison is made between DeepSeek and another model here.
Follow the lab
New test reports and tier-list changes get announced in my Discord's #news channel. The full videos are on YouTube.
Receipts
DanManREAL · AI lab
Test receipt
DeepSeek V4.1 Flash
- DeepSWE v1.1
- 74.2%DeepSeek vendor comparison table (model publisher). Original V4.1 Flash, not Flash-1017; Mini-SWE-V2. Not independently rerun by DanManREAL.
- Terminal-Bench 4.0
- 31.2%DeepSeek vendor comparison table (model publisher). Original V4.1 Flash, not Flash-1017; Claude Code harness. Not independently rerun by DanManREAL.
- ProgramBench
- 20.3DeepSeek vendor comparison table (model publisher). Original V4.1 Flash; metric as published. Not independently rerun by DanManREAL.
- Source review
- November 17, 2026Date these source claims were reviewed; not a claim about when the published benchmark was executed.
Verdict
Published benchmark evidence
Tier list The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
danmanreal.com/tests/deepseek-v4-1-flash-public-benchmarks
Models covered by external benchmarks
On the tier list's bench
- DeepSeek V4.1 Flash · 5 receipts on file
Related
- AI Coding Agents Tier List
- My real DeepSeek bill: 12.31 billion tokens for $122.28
- Does an extra AI review step pay off? A small DeepSeek trial said no
Independent testing. Sponsorships are always disclosed, and verdicts are never for sale.
Disagree? Bring your receiptsAll reports
- One inbox for my AI coding agents: triage took 34% less time, resuming work did not improveHead-to-headNov 20
- Mythos 5.2: research and security benchmark source reviewPublic benchmarkNov 17
- Fable 5.2: agentic coding benchmark source reviewPublic benchmarkNov 17
- Claude Opus 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- Astra 6.2: new model and public task-evaluation evidencePublic benchmarkNov 17
- Sol 6.2: new model and public coding benchmark evidencePublic benchmarkNov 17
- Gemini 4 Carbon: benchmark source and usage profilePublic benchmarkNov 17
- Qwen 4: Qwen4-Max launch and benchmark sourcesPublic benchmarkNov 17
- Kimi K3.1 Pro: public coding benchmark receiptsPublic benchmarkNov 17
- GLM-5.3: public coding benchmark receiptsPublic benchmarkNov 17
- Gemini 4 Argon: public coding benchmark receiptsPublic benchmarkNov 17
- Claude Sonnet 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- DeepSeek V4.1 Flash: public coding benchmark receiptsPublic benchmarkNov 17
- Muse Wonder Full 1.0: public coding benchmark receiptsPublic benchmarkNov 17
- Grok 4.7: separate release and public coding resultsPublic benchmarkNov 17
- ai-design-check on two of my FABLE screens: 10 warnings, 4 of them false alarmsField testNov 2
- Vercel Agent Desktop used a sample Trip app in 6 actions without moving the mouseField testNov 1
- Cerberus Guard on sample data: one crawler name, two verdicts, and a block earned by behaviorField testOct 31
- GPT-6.1 Terra vs Opus 5.5: three real projects, and the designs I picked blindHead-to-headOct 29
- Steps → Reduced: opening a terminal agent went from 98 seconds to 11AuditOct 28
- One ChatGPT chat directed Opus and Codex at once through DMR_REMOTEField testOct 28
- Claude Duet ran my channel for a week and caught 269 fake AI crawlersField testOct 27
- Kimi K3.1 Pro built my animation family, after a quota stop and 21 minutes of failed requestsField testOct 24
- ChatGPT + Codex built 33 new DMR_REMOTE toolsField testOct 23
- Gemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistencyField testOct 21
- Qwen 3.8 Flash Next on my RTX 5080: 81 tokens a second with 202K tokens readAuditOct 21
- Does an extra AI review step pay off? A small DeepSeek trial said noHead-to-headOct 19
- My real DeepSeek bill: 12.31 billion tokens for $122.28Cost receiptOct 17
- Gemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeoutIncidentOct 14
- I asked 10 AI coding agents for their biggest weaknessSelf-reportOct 14
- Jarvis 27B on my own PC: 30 tokens a second, and it still recalls at 251KAuditOct 10
- My agents' first game benchmark was wrong. Here's how it was caughtIncidentOct 10
- Pixal3D + Blender turned my 2D art into animated 3D charactersField testSep 22
- DeepSeek Flash answered 200 OK and sent nothing for two hoursIncidentSep 14
- Grok, Codex and DeepSeek played WorldOS for Rank 1. Grok won with 98Head-to-headAug 20
- DeepSeek V4 Flash audited 18 of my real coding sessions. Two said 'done' too earlyAuditAug 4