Up to date
3.82B parameters, text in, text out, 4K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #139 in LLMs · score 93.8 method v1 Sources 3 Downloads 30d 340.8k · #206 of 619 tracked
Changed this week: 2 new quantized builds on the Hub (183 in all).2026-10-24
Picked by rule from 27 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Phi-3 models achieve high performance rivaling Mixtral 8x7B by training on a scaled-up dataset of heavily filtered publicly available web data and synthetic data.
Microsoft's Phi-3-Mini is a 3.8 billion-parameter language model designed for efficiency on devices with limited resources like notebooks and smartphones.
The 2 picks above, plus these 23. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
| # | Cheapest live row | In /1M | Out /1M | Context | Quant | Seen |
|---|---|---|---|---|---|---|
| 1 | Featherless | — | — | — | — | Oct 25 |
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.
Served by 1 API provider; heavily run locally (183 quantized builds); a big base for builders (2,006 fine-tunes, adapters and merges).
Top Spaces running it: ginipick/FLUXllama ♥ 561 · KingNish/JARVIS ♥ 292 · Nymbo/Serverless-TextGen-Hub ♥ 44 · MBZUAI/LLaMA-3-V ♥ 33
Weights only, an estimate: leave room for the KV cache. Method
About 2.3 GB at INT4, 4.6 GB at FP8, 9.2 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 (2024-04-22), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
2,189 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Average | 36.7 | June 2024 Update | model author | 2026-10-07 |
| author-reported GPQA Diamond · Reason through hard problems | 30.6 | June 2024 Update | model author | 2026-10-07 |
| author-reported Instruction Extra Hard | 6 | June 2024 Update | model author | 2026-10-07 |
| author-reported Instruction Hard | 5.1 | June 2024 Update | model author | 2026-10-07 |
| author-reported Instructions Challenge | 42.3 | June 2024 Update | model author | 2026-10-07 |
| author-reported JSON Structure Output | 52.3 | June 2024 Update | model author | 2026-10-07 |
| author-reported MMLU · Reason through hard problems | 70.9 | June 2024 Update | model author | 2026-10-07 |
| author-reported XML Structure Output | 49.8 | June 2024 Update | model author | 2026-10-07 |
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 LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | Phi3ForCausalLM |
| Model type | phi3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,821,079,552 |
| Size band | M |
| MoE | no |
| Context length | 4,096 (config.json max_position_embeddings) |
| Languages | en, fr |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Phi3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Phi3ForCausalLM · 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 Phi-3-mini-4k-instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 9.2 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 4.6 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 2.3 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 183finetune: 1,124adapter: 869merge: 13
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| solidrust/Phi-3-mini-4k-instruct-AWQ | quantized | AWQ | 3.82B | M | — | solidrust | 0 | 645.5k |
| mlc-ai/Phi-3-mini-4k-instruct-q4f16_1-MLC | quantized | — | — | — | — | mlc-ai | 1 | 209.6k |
| bartowski/Phi-3.1-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | bartowski | 45 | 163.4k |
| mozilla-ai/Phi-3-mini-4k-instruct-llamafile | finetune | — | — | — | apache-2.0 | mozilla-ai | 16 | 108.2k |
| e-palmisano/Phi3-ITA-mini-4K-instruct | finetune | — | 3.82B | M | mit | e-palmisano | 1 | 96.1k |
| lmstudio-community/Phi-3.1-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | lmstudio-community | 24 | 76.6k |
| OpenVINO/Phi-3-mini-4k-instruct-int4-ov | quantized | int4 | — | — | mit | OpenVINO | 4 | 72.9k |
| numind/NuExtract | finetune | — | 3.82B | M | mit | numind | 236 | 43.6k |
| qwp4w3hyb/Phi-3-mini-4k-instruct-iMat-GGUF | quantized | GGUF | — | — | mit | qwp4w3hyb | 1 | 40.8k |
| second-state/Phi-3-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | second-state | 5 | 40.1k |
| gaianet/Phi-3-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | gaianet | 1 | 36.4k |
| legraphista/Phi-3-mini-4k-instruct-update2024_07_03-IMat-GGUF | quantized | GGUF | — | — | mit | legraphista | 0 | 32.9k |
| ycheng1024/task-20-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | ycheng1024 | 0 | 32.9k |
| bartowski/Phi-3-mini-4k-instruct-v0.3-GGUF | quantized | GGUF | — | — | mit | bartowski | 5 | 25.5k |
| MaziyarPanahi/calme-2.3-phi3-4b | finetune | — | 3.82B | M | mit | MaziyarPanahi | 9 | 23.9k |
| brittlewis12/Phi-3-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | brittlewis12 | 2 | 22.1k |
| legraphista/Phi-3-mini-4k-instruct-IMat-GGUF | quantized | GGUF | — | — | mit | legraphista | 0 | 21.9k |
| ContinuousAT/Phi-CAT | adapter | — | — | — | — | ContinuousAT | 0 | 20.3k |
| backyardai/Phi-3-mini-4k-instruct-GGUF | quantized | GGUF | — | — | mit | backyardai | 0 | 16.7k |
| ab-ai/PII-Model-Phi3-Mini | finetune | — | 3.82B | M | mit | ab-ai | 10 | 14.1k |
| QuantFactory/NuExtract-GGUF | quantized | GGUF | — | — | mit | QuantFactory | 4 | 9.5k |
| flockgo/task-1-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | flockgo | 0 | 8.8k |
| DreamGallery/task-5-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | DreamGallery | 0 | 8.3k |
| lucifer-ms/task-3-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | lucifer-ms | 0 | 8.3k |
| DreamGallery/task-1-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | DreamGallery | 0 | 7.5k |
| KrithikV/MedMobile | finetune | — | 3.82B | M | mit | KrithikV | 8 | 6.9k |
| mlc-ai/Phi-3-mini-4k-instruct-q4f32_1-MLC | quantized | — | — | — | — | mlc-ai | 0 | 7.3k |
| matrixportalx/Phi-3-mini-4k-instruct-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | matrixportalx | 4 | 6.3k |
| moelanoby/phi-3-M3-coder | finetune | — | 3.82B | M | — | moelanoby | 48 | 4k |
| sam-fd/task-13-microsoft-Phi-3-mini-4k-instruct | adapter | — | 3.82B | M | — | sam-fd | 0 | 6.2k |
| DreamGallery/task-6-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | DreamGallery | 0 | 6k |
| latiao1999/task-5-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | latiao1999 | 0 | 6k |
| QuantFactory/phi-3-portuguese-tom-cat-4k-instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | QuantFactory | 2 | 5.8k |
| Phora68/bible-study-phi3-mini | quantized | GGUF | 3.82B | M | mit | Phora68 | 3 | 5.4k |
| ishanmane/phi3-rewriter-sft | adapter | — | — | — | — | ishanmane | 0 | 5.5k |
| misterJB/arkadas-field-717hz | finetune | — | 3.82B | M | — | misterJB | 0 | 5.5k |
| srrthk/task-13-microsoft-Phi-3-mini-4k-instruct | adapter | — | — | — | — | srrthk | 0 | 4.9k |
| leliuga/Phi-3-mini-4k-instruct-bnb-4bit | quantized | -bnb | 3.93B | M | other | leliuga | 6 | 4.5k |
| QuantFactory/calme-2.3-phi3-4b-GGUF | quantized | GGUF | — | — | mit | QuantFactory | 0 | 4.5k |
| edumunozsala/phi-3-mini-QLoRA | adapter | — | — | — | mit | edumunozsala | 0 | 4.2k |
| Date | Commit | SHA |
|---|---|---|
| 2025-12-10 | Update README.md (#109) | f39ac1d |
| 2025-12-10 | Upload data_summary_card.md (#108) | 87e6c1e |
| 2024-09-20 | Update README.md (#96) | 0a67737 |
| 2024-08-20 | Update README.md | 5a516f8 |
| 2024-07-01 | Minor updates on README | c1358f8 |
| 2024-07-01 | Update model (#77) | 4f818b1 |
| 2024-06-11 | Add attention_bias to make TGI work (#64) | ff07dc0 |
| 2024-05-31 | Add link to Phi-3 vision ONNX models | d269012 |
| 2024-05-22 | Update sample_finetune.py (#65) | 5fa3419 |
| 2024-05-21 | Update README | 19f7775 |
| 2024-05-20 | Update README.md | d6b5c7b |
| 2024-05-16 | Fix typo (#58) | 8f5f3a0 |
| 2024-05-03 | fix(README): Adds temperature=0.7 to prevent Inference API from going off the chart. | 920b6cf |
| 2024-05-02 | fix(tokenizer): Also updates tokenizer.json to prevent any mismatch. | 653ee82 |
| 2024-05-02 | fix(tokenizer_config): Adjusts `rstrip` of special tokens. (#53) | 4eea1a7 |
| 2024-04-29 | Update README.md | 2921602 |
| 2024-04-29 | fix(root): Replaces system by user to improve generation experience. | 300945e |
| 2024-04-28 | Update docstrings | 291e9e3 |
| 2024-04-28 | Update docstring in sample_finetune.py | 13565ec |
| 2024-04-26 | Update README.md | cc80d6c |
| 2024-04-26 | Update sample_finetune.py (#42) | 6a07688 |
| 2024-04-26 | chore(readme): Adds a proper chat example to widget. | 4b6011c |
| 2024-04-25 | fix(readme): Adds information about placeholder tokens. | 3c0c9df |
| 2024-04-22 | chore(root): Initial files upload. | 65be4e0 |
| 2024-04-22 | initial commit | c91d28f |
Hosted endpoints observed for Phi-3-mini-4k-instruct, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Featherless hf | microsoft/Phi-3-mini-4k-instruct | — | — | — | — | — | — | — | live | 2026-10-25 |
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 21:35 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:02 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.
