DeepSWE
No official DeepSWE mini-swe-agent row for this identity on the public board we fetched.
Independent lab. Official Apache-2.0 repo. Not on the week board we cite. AA title is “Inkling (xhigh)”; Intelligence Index 42. vals hero is 34.10% ±1.13, not the 47.57% Terminal-Bench Updates sentence. No official DeepSWE row.
Not listed as a Continuum hosted id. Apache-2.0 on the official repo. OpenRouter catalog created unix 1784325956 (17 Jul 2026). 1,048,576 in / 262,144 out; text + image + audio in, text out. Hugging Face id thinkingmachines/Inkling.
HF config.json (fetched 19 Aug 2026): InklingForConditionalGeneration, model_type inkling_mm_model. Text stack: 66 layers, hidden 6144, 256 routed experts, 6 experts/token, 2 shared experts. Vision config present. Chat template ships thinking-effort tokens.
Use Inkling when you want this official open checkpoint. AA 42 is the xhigh tile; vals hero is 34.10% ±1.13. We do not invent a DeepSWE percent and we do not promote the 47.57% Terminal-Bench Updates sentence to the hero.
No official DeepSWE mini-swe-agent row for this identity on the public board we fetched.
vals.ai Inkling card, fetched 19 Aug 2026: hero Index 34.10% ±1.13, $1.354/test, 19 min 37 s. Accuracy Rankings bars printed 0.0% placeholders: omitted. Updates prose prints SWE-bench Verified subset 75.49%, CorpFin v2 subset 69.23%, Finance Agent v2 45.97%, Vibe Code subset 13.28%, and Terminal-Bench 2.1 47.57% across three full trials. Recommended eval settings on that page: temperature 1, top-p 1, up to 256k output tokens, separate reasoning enabled, reasoning effort set to "0.99". Those named percents are not the hero chip.
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.
Copied from AA or vals HTML we opened. 0.0% placeholder bars are omitted. Do not average these into the hero tiles, and do not invent CI, $/task, or steps for Artificial Analysis.
| Bench | Printed | Note | Source | URL | As of |
|---|---|---|---|---|---|
| AA list pair (xhigh tile) | $1.00 / $4.05 per 1M | Printed AA In/Out. OpenRouter catalog is $0.95 / $4.05 plus $0.16/M cache read. | Artificial Analysis Inkling | source | 2026-08-19 |
| AA throughput | 74.5 tok/s | Printed Speed row. Not a coding score. | Artificial Analysis Inkling | source | 2026-08-19 |
| AA Index eval cost | $698.28 | Printed “it cost $698.28 to evaluate Inkling (xhigh) on the Intelligence Index.” | Artificial Analysis Inkling | source | 2026-08-19 |
| vals Index extras | $1.354/test · 19 min 37 s | Printed Cost / Test and Latency on the vals hero. Hero Accuracy is 34.10% ±1.13. | vals.ai Inkling | source | 2026-08-19 |
| vals Updates SWE-bench Verified subset | 75.49% | Updates prose, Vals Index subset. Not DeepSWE official and not the hero. | vals.ai Inkling | source | 2026-08-19 |
| vals Updates CorpFin v2 subset | 69.23% | Updates prose, Vals Index subset. | vals.ai Inkling | source | 2026-08-19 |
| vals Updates Finance Agent v2 | 45.97% | Updates prose, Index subset. | vals.ai Inkling | source | 2026-08-19 |
| vals Updates Vibe Code subset | 13.28% | Updates prose, Vals Index subset. | vals.ai Inkling | source | 2026-08-19 |
| vals Updates Terminal-Bench 2.1 | 47.57% | Updates prose: “47.57% across three full trials.” Not the hero chip. | vals.ai Inkling | source | 2026-08-19 |
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.
Community quants: Community quants are community.
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-00108.safetensors | 19,981,728,408 |
model-00002-of-00108.safetensors | 10,551,335,256 |
model-00003-of-00108.safetensors | 19,793,037,628 |
model-00004-of-00108.safetensors | 19,554,020,792 |
model-00005-of-00108.safetensors | 19,402,851,712 |
model-00006-of-00108.safetensors | 19,632,630,372 |
model-00007-of-00108.safetensors | 19,528,890,012 |
model-00008-of-00108.safetensors | 19,327,352,992 |
model-00009-of-00108.safetensors | 9,817,920,224 |
model-00010-of-00108.safetensors | 19,579,378,432 |
model-00011-of-00108.safetensors | 19,985,018,028 |
model-00012-of-00108.safetensors | 19,327,418,768 |
Showing the first 12 of 109 safetensor shards. The tree also holds 16 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 97 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.
Not listed as a Continuum hosted id. OpenRouter thinkingmachines/inkling.
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.
OpenRouter catalog 19 Aug 2026: $0.95 / $4.05 per million, cache read $0.16 /M, 1,048,576 / 262,144, created 1784325956. No DeepSWE $/task.
Artificial Analysis title is “Inkling (xhigh)”. That page prints In $1.00 / Out $4.05, 74.5 tok/s, and “it cost $698.28 to evaluate Inkling (xhigh) on the Intelligence Index.” vals prints $1.354 per Index test and 19 min 37 s.
Artificial Analysis title is “Inkling (xhigh)”. Intelligence Index 42, fetched 19 Aug 2026. Same page prints $1.00 / $4.05, 74.5 tok/s, and $698.28 to run the Index. The AA tile is the published integer 42. AA did not print a DeepSWE-style CI or step count we will copy onto the chip.
| Claim | Source | As of |
|---|---|---|
| OR catalog $0.95 / $4.05, cache read $0.16/M, 1,048,576 / 262,144, created 1784325956, hugging_face_id thinkingmachines/Inkling | OpenRouter catalog | 2026-08-19 |
| HF safetensors.total 952,377,623,626; usedStorage 1,909,218,968,769; 109 safetensor files; likes 1,736; downloads 125,873; created 2026-07-14T13:23:14Z; Apache-2.0 | Hugging Face API thinkingmachines/Inkling | 2026-08-19 |
| AA title Inkling (xhigh); Index 42; $1.00 / $4.05; 74.5 tok/s; $698.28 to evaluate Index | Artificial Analysis Inkling | 2026-08-19 |
| vals hero 34.10% ±1.13; $1.354/test; 19 min 37 s; Updates subset benches; settings temperature 1, top-p 1, 256k out, reasoning effort "0.99" | vals.ai Inkling | 2026-08-19 |
| OpenRouter id thinkingmachines/inkling, context 1,048,576 | OpenRouter /api/v1/models | 2026-08-19 |
| HF downloads 125,873 | Hugging Face API thinkingmachines/Inkling | 2026-08-19 |
| 109 safetensor shards; shard bytes 1,904,755,463,940; tree files 125 | Hugging Face tree API thinkingmachines/Inkling | 2026-08-19 |
| Spaces API returned 19; official collection Inkling (4 items, 58 upvotes) | Hugging Face Spaces / collections API | 2026-08-19 |
No other card in this cluster shares a published DeepSWE official row we can put next to this one. We will not compare on vendor-blog numbers.
Continue through Thinking Machines's model family with Inkling Small.
The official DeepSWE board updated 13 Aug 2026 did not list Inkling. Omitting is the rule. vals’ SWE-bench Verified subset 75.49% is a different harness and a subset.
That number is Updates prose about Terminal-Bench 2.1 across three full trials. The vals hero is 34.10% ±1.13.
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.
No Continuum hosted id is verified for this model, so this page does not invent a curl target. Use the first-party API or the OpenRouter slug thinkingmachines/inkling.