DeepSWE
No official DeepSWE mini-swe-agent row for this identity on the public board we fetched.
Context window on the 2026-08-19 OpenRouter row: 131,072 tokens.
Max completion tokens on that row: 131,072.
Input / output on that row: $0.05 / $0.33 per 1M tokens.
Architecture fields: text->text · Llama3 · llama3.
Hugging Face id on that row: meta-llama/Llama-3.2-3B-Instruct.
Llama 3.2 3B is a 3-billion-parameter multilingual large language model, optimized for advanced natural language processing tasks like dialogue generation, reasoning, and summarization.
When you want the 2024-09-25 catalog SKU, not a later rename.
No independent board lists this exact SKU as of 2026-08-19.
Closest benchmarked sibling: Muse Spark 1.2, 55% ±2% Pass@1 at xhigh on the DeepSWE official board. That is a score for the sibling, not for this SKU.
Boards checked: DeepSWE official (2026-08-13), Artificial Analysis Intelligence Index, vals.ai hero index. Every card that does carry a row is on the coding leaderboard.
No official DeepSWE mini-swe-agent row for this identity on the public board we fetched.
No Intelligence Index integer fetched for this identity. Chip omitted.
We did not open a vals.ai card for this identity. Chip omitted.
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.
Official weights id on the 2026-08-19 OpenRouter row: meta-llama/Llama-3.2-3B-Instruct. Community GGUF is a quant, not a second Continuum card.
Hugging Face Hub API 2026-08-19: gated=manual.
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-00002.safetensors | 4,965,799,096 |
model-00002-of-00002.safetensors | 1,459,729,952 |
original/consolidated.00.pth | 6,425,585,114 |
tokenizer.json | 9,085,657 |
original/tokenizer.model | 2,183,982 |
tokenizer_config.json | 54,528 |
README.md | 41,744 |
model.safetensors.index.json | 20,919 |
LICENSE.txt | 7,712 |
USE_POLICY.md | 6,021 |
.gitattributes | 1,519 |
config.json | 878 |
special_tokens_map.json | 296 |
original/orig_params.json | 220 |
original/params.json | 220 |
generation_config.json | 189 |
All 16 files the tree API returned are listed: every one of the 2 safetensor shards plus 14 other files. Files tab: https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct/tree/main.
OpenRouter slug meta-llama/llama-3.2-3b-instruct on the 2026-08-19 catalog.
First party: https://www.llama.com/.
Not on the Continuum host list fetched 2026-08-19.
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 2026-08-19: $0.05 in / $0.33 out per 1M tokens. Context 131,072 in / 131,072 out.
No separate internal-reasoning price on that row.
| Claim | Source | As of |
|---|---|---|
| OpenRouter id meta-llama/llama-3.2-3b-instruct; context 131072; created 1727222400; $0.05 / $0.33 per 1M | OpenRouter /api/v1/models | 2026-08-19 |
| OpenRouter id meta-llama/llama-3.2-3b-instruct, context 131,072 | OpenRouter /api/v1/models | 2026-08-19 |
| HF downloads 1,102,483 | Hugging Face API meta-llama/Llama-3.2-3B-Instruct | 2026-08-19 |
| HF safetensors.total: 3,212,749,824 stored tensors; HF API usedStorage 12,853,298,144 bytes; created 2024-09-18T15:19:20.000Z; HF tensors BF16 | Hugging Face API meta-llama/Llama-3.2-3B-Instruct | 2026-08-19 |
| 2 safetensor shards; shard bytes 6,425,529,048; tree files 16 | Hugging Face tree API meta-llama/Llama-3.2-3B-Instruct | 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 Meta's model family with Llama 3.1 70B Instruct.
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 meta-llama/llama-3.2-3b-instruct.