Up to date
thegovind Research / non-commercial (blink-research) out 31 days ago
26.9B parameters, text in, labels out, 256K tokens of context.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 144 · #615 of 619 tracked
Picked by rule from 2 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Blink is a terminal application for iOS and iPadOS that provides a distraction-free command line experience.
The 1 pick above, plus these 1. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 2 Spaces.
Weights only, an estimate: leave room for the KV cache. Method
About 16 GB at INT4, 32 GB at FP8, 65 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-09-24), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Computer use in ten apps | 87% percent | Overall | model author | 2026-10-07 |
| author-reported Computer use in ten apps | 30/30 | Shop, settings, files | model author | 2026-10-07 |
| author-reported Computer use in ten apps | 20/20 | Calendar, phone | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 52.72 | Local development readout | model author | 2026-10-07 |
| author-reported JevBench public hard items | 90/111 | batched | model author | 2026-10-07 |
| author-reported JevBench public hard items | 89/111 | via serial `serve.py` | model author | 2026-10-07 |
The model provides probabilities for choice, noul, or score questions without generating text.[1]
Generated 2026-10-07 15:19 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in Decision models
| Pipeline tag | text-classification |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5ForCausalLM |
| Model type | qwen3_5_text |
| Dtype | bfloat16 |
| Params by dtype | BF16: 26,895,998,464 |
| Size band | L |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| Languages | en |
| License | Research / non-commercial (blink-research) (other) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3_5ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for blink-27b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 65 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 32 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 16 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
| How it was made | SFT on target distributions (program-generated, public, teacher-written rows), adapters merged |
|---|---|
| Lineage bucket | F - fine-tune of an open model (public/synthetic data) |
| Lineage evidence | model card + HF base_model |
| Base model | Qwen/Qwen3.5-4B ; Qwen/Qwen3.8-27B |
| DI 0.2.1 score | 52.12 / 63.44 (DI 0.1 edition) (self-reported (older edition)) |
| Hosted price (input) | self-hosted/M |
| Source | https://huggingface.co/thegovind/blink-27b |
Base model: Qwen/Qwen3.8-27B
| Date | Commit | SHA |
|---|---|---|
| 2026-09-30 | Card: customize it | 34f301c |
| 2026-09-30 | Card: run it on a Mac with MLX | 198d250 |
| 2026-09-30 | Card: computer-use Code link points at examples/cua | 61a6f75 |
| 2026-09-29 | Card: crisper card, computer use first | a8c7379 |
| 2026-09-28 | Card: crisper card, computer use first | 4f6f817 |
| 2026-09-27 | Card: link Screen click and the Computer use docs | aec170f |
| 2026-09-27 | Code revision v1.4: card update | 59ce811 |
| 2026-09-27 | Code revision v1.3: opt-in screenshot input (self-hosted) | e3d6f9b |
| 2026-09-26 | Shorter card; link the API docs | badd4ad |
| 2026-09-26 | Card: links to the source code and docs | 275ef8b |
| 2026-09-26 | v1.2: TypeSafe API wire compatibility in serve.py (server-side SDKs, GET /v1/models, optional API key, error detail, request ids; a full batching queue answers 529); score answers carry confidence; weights unchanged | 1f18127 |
| 2026-09-26 | Card: local Decision Index 0.2 run (descriptive; known training exposure not penalized) and its evidence file | 69611a5 |
| 2026-09-25 | Card: architecture, post-training and readout diagrams | 3dd3c9b |
| 2026-09-24 | Card and licence corrections: audit scope, reproduction wording, MiMo lineage | 9e7dfc2 |
| 2026-09-24 | blink v1.0 | 295bab7 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:44 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:19 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
