DeepSWEmini-swe-agent
53%±4%
- bench
- DeepSWE
- version
- v1.1
- split
- public 113 tasks
- harness
- mini-swe-agent
- effort
- max
- n
- 113
- metric
- Pass@1
- value
- 53%
- ci
- ±4%
- $/task
- $0.10
- tokens
- 108k out
- steps
- 153
deepseek-v4-flashCurrent weights 0731. Unversioned deepseek/deepseek-v4-flash is 0423 lineage. Continuum hosted id is the bare deepseek-v4-flash.
HF config.json (fetched 19 Aug 2026): DeepseekV4ForCausalLM, 43 layers, hidden 4096, 256 routed experts, 6 experts/token, 1 shared expert, YaRN to 1,048,576, DSpark attached (dspark_block_size 5). Official DeepSeek docs list context 1M and max out 384K. OpenRouter’s 0731 row is 1,310,720 / 393,216. Those are two published windows; do not average them.
No Jinja chat template: the repo ships an encoding/ folder (same pattern as Pro 0813).
Use Flash for volume, for hosted DeepSeek on Continuum, and when $0.10 DeepSWE tasks matter more than the top pass rate. Official board: 53% Pass@1 ±4% at max, $0.10 per task, 108k output, 153 steps.
Ranked view: Best coding models: the independent leaderboard puts this row and every other card that carries an independent coding score on one board, with the confidence intervals left visible.
Plus is $25/mo with $25 weekly hosted usage. The Mac app stays free with your own keys. Get Plus is not Download for Mac.
Preview / prior weights: Prior weights: DeepSeek-V4-Flash (0423 lineage).
Community quants: Any GGUF or AWQ you find under other orgs is community, not official.
DSpark speculative decoding ships in this checkpoint too (dspark_block_size 5). That is a config field, not a second repo.
Official DeepSeek API docs (fetched 19 Aug 2026) put Flash and Pro on one page with two price columns. Flash is the left column. Copying Pro cents onto Flash is a fabricated schedule.
This is the model repo, not Download for Mac. Downloads and likes from the Hugging Face API on 2026-08-19.
usedStorage and the sum of .safetensors sizes can differ (LFS pointers, non-shard files, duplicate copies). Cited separately. Source: recursive tree API, 2026-08-19.
| Path | Bytes |
|---|---|
model-00001-of-00048.safetensors | 1,059,061,856 |
model-00002-of-00048.safetensors | 3,566,321,192 |
model-00003-of-00048.safetensors | 3,566,321,192 |
model-00004-of-00048.safetensors | 3,596,229,272 |
model-00005-of-00048.safetensors | 3,568,768,976 |
model-00006-of-00048.safetensors | 3,590,024,776 |
model-00007-of-00048.safetensors | 3,568,768,976 |
model-00008-of-00048.safetensors | 3,590,024,776 |
model-00009-of-00048.safetensors | 3,568,768,976 |
model-00010-of-00048.safetensors | 3,590,024,776 |
model-00011-of-00048.safetensors | 3,568,768,976 |
model-00012-of-00048.safetensors | 3,590,026,352 |
Showing the first 12 of 48 safetensor shards. The tree also holds 26 non-shard files, none of them listed here. The snapshot was capped at 12 shards when the tree API was read on 2026-08-19, so the remaining 36 are on the Files tab, not on this page. Totals above are the full tree API sum, not this slice. Byte sizes are Hub tree API size fields, never invented.
Continuum hosts deepseek-v4-flash (19 Aug 2026 public allowlist).
First-party DeepSeek API model id on the official pricing page is deepseek-v4-flash, version DeepSeek-V4-Flash-0731. OpenRouter slug is deepseek/deepseek-v4-flash-0731.
deepseek-v4-flash
click to select
We list first-party and OpenRouter, plus Continuum only when the live public allowlist named the id. This is not a 15-host routing table.
DeepSWE official: $0.10 per task at max (108k out, 153 steps). That is the hero economics number.
First-party DeepSeek API docs, fetched 19 Aug 2026: one USD schedule with peak and off-peak. Peak hours 01:00–04:00 and 06:00–10:00 UTC; all other hours off-peak at half. Flash cache-hit / cache-miss / output: peak $0.014 / $0.44 / $1.32 per million; off-peak $0.007 / $0.22 / $0.66. Concurrency cap 2500 (Pro is 500). These are not the Pro column ($0.044 / $1.32 / $3.96 peak).
OpenRouter catalog 19 Aug 2026 for deepseek-v4-flash-0731 lists $0.14 / $0.28. Continuum’s hosted rate card matches those cents. That catalog pair is not the official DeepSeek Flash peak/off-peak table. Artificial Analysis printed the official Flash peak miss/out pair ($0.44 / $1.32). Cite all three; do not collapse them into one mystery number.
Artificial Analysis Intelligence Index 52, fetched 19 Aug 2026. AA printed $0.44 / $1.32 for Flash: that is the official DeepSeek peak cache-miss / output pair, not a fourth table. AA did not publish a DeepSWE-style CI, step count, or $/task on the chip we will copy; those fields stay omitted on the AA tile. Do not average AA cost with DeepSWE $0.10/task.
| Claim | Source | As of |
|---|---|---|
| DeepSWE 53% ±4% at max | DeepSWE official board | 2026-08-13 |
| Official Flash peak $0.014 / $0.44 / $1.32 and off-peak $0.007 / $0.22 / $0.66; peak 01:00–04:00 and 06:00–10:00 UTC; Flash concurrency 2500 | DeepSeek API Models & Pricing | 2026-08-19 |
| HF safetensors.total 304,180,418,494; usedStorage 166,888,735,421 bytes; 48 shards; likes 3,545; created 2026-07-31T07:30:24Z; arXiv:2606.19348 | Hugging Face API DeepSeek-V4-Flash-0731 | 2026-08-19 |
| config DeepseekV4ForCausalLM, 43 layers, 256 routed experts, 6/tok, DSpark 5 | HF config.json DeepSeek-V4-Flash-0731 | 2026-08-19 |
| OR catalog $0.14 / $0.28 for flash-0731; AA printed $0.44 / $1.32 | OpenRouter catalog + Artificial Analysis DeepSeek V4 Flash | 2026-08-19 |
| Hosted id deepseek-v4-flash | GET /v1/chat/hosted/models/public | 2026-08-19 |
| 11.3T week tokens on flash-0731 | OpenRouter rankings This Week | 2026-08-19 |
| OpenRouter id deepseek/deepseek-v4-flash-0731, context 1,310,720 | OpenRouter /api/v1/models | 2026-08-19 |
| HF downloads 2,330,940 | Hugging Face API deepseek-ai/DeepSeek-V4-Flash-0731 | 2026-08-19 |
| 48 safetensor shards; shard bytes 166,886,535,336; tree files 74 | Hugging Face tree API deepseek-ai/DeepSeek-V4-Flash-0731 | 2026-08-19 |
| Spaces API returned 22; official collection DeepSeek-V4 (4 items, 833 upvotes) | Hugging Face Spaces / collections API | 2026-08-19 |
Same official DeepSWE harness (mini-swe-agent, public 113 tasks), fetched 2026-08-19. Different effort labels are the lab's own setting on that board, shown here rather than normalized.
Continue through DeepSeek's model family with R1.
Continuum lists deepseek-v4-flash without a date suffix. Official DeepSeek docs name the current version DeepSeek-V4-Flash-0731. We do not invent a silent pin beyond that.
No. Official docs: Flash off-peak $0.007 / $0.22 / $0.66; Pro off-peak $0.022 / $0.66 / $1.98. Same page, two columns.
Those are different published tables. Official DeepSeek is peak/off-peak USD. OpenRouter’s 0731 catalog row is $0.14 / $0.28. AA printed official peak miss/out. We cite each source; we do not pick a winner.
Lab posts pick a harness, an effort, a split, and sometimes a private eval set. DeepSWE publishes the official mini-swe-agent row with cost, tokens, and steps. Scale SWE-bench Pro only counts when the public shared-harness board has a row we can fetch. A higher lab number is usually a different test, not a better one.
Verified against the live public allowlist on 2026-08-19. Base URL is https://continuumcode.ai/v1. Keys are cont_sk_ from Settings, Account, Inference API. Personal keys need Plus or above.
curl https://continuumcode.ai/v1/chat/completions \
-H "Authorization: Bearer $CONTINUUM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"Review this diff."}]}'