Up to date
microsoft MIT out 18 months ago
3.84B parameters, text in, text out, 128K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #494 in LLMs · score 76.8 method v1 Sources 3 Downloads 30d 38.6k · #400 of 619 tracked
Picked by rule from 17 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Phi-4 is a 14B parameter open model designed for memory/compute constrained environments and latency bound scenarios.
Phi-4-reasoning-vision-15B is a 15 billion parameter open-weight multimodal reasoning model trained with just 200 billion tokens of multimodal data.
The 2 picks above, plus these 15. 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; 44 quantized builds for local use.
Top Spaces running it: KavinduHansaka/AI-Powered_Disease_Diagnostic_Assistant ♥ 2
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 (2025-04-29), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
81 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 AIME 2024 · Reason through hard problems | 57.5 | Phi-4-mini-reasoning (3.8B) | model author | 2026-10-07 |
| author-reported GPQA Diamond · Reason through hard problems | 52 | Phi-4-mini-reasoning (3.8B) | model author | 2026-10-07 |
| author-reported MATH-500 · Reason through hard problems | 94.6 | Phi-4-mini-reasoning (3.8B) | model author | 2026-10-07 |
Phi-4-mini-reasoning is a 3.84B parameter transformer-based language model developed by microsoft for mathematical reasoning tasks.[1][2][3] The model supports a 128K token context length and is released under the MIT license.[4][5]
Generated 2026-10-07 15:11 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 LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | Phi3ForCausalLM |
| Model type | phi3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,836,021,760 |
| Size band | M |
| MoE | no |
| Context length | 131,072 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2504.21233 |
| 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-4-mini-reasoning: 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: 44finetune: 26merge: 1adapter: 10
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| lmstudio-community/Phi-4-mini-reasoning-MLX-4bit | quantized | MLX | 3.84B | M | mit | lmstudio-community | 5 | 986.3k |
| lmstudio-community/Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | lmstudio-community | 6 | 218.6k |
| unsloth/Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | unsloth | 81 | 177.2k |
| unsloth/Phi-4-mini-reasoning-unsloth-bnb-4bit | quantized | -bnb | 3.93B | M | mit | unsloth | 9 | 88.5k |
| bartowski/microsoft_Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | bartowski | 17 | 41.3k |
| unsloth/Phi-4-mini-reasoning | finetune | — | 3.84B | M | mit | unsloth | 6 | 16.9k |
| unsloth/Phi-4-mini-reasoning-bnb-4bit | quantized | -bnb | 3.94B | M | mit | unsloth | 1 | 8.2k |
| mlx-community/Phi-4-mini-reasoning-4bit | quantized | -4bit | 3.84B | M | mit | mlx-community | 0 | 7.5k |
| mradermacher/Phi-4-mini-reasoning-i1-GGUF | quantized | GGUF | — | — | mit | mradermacher | 1 | 7k |
| mradermacher/Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | mradermacher | 0 | 2.9k |
| litert-community/Phi-4-mini-reasoning | quantized | — | — | — | mit | litert-community | 4 | 2.2k |
| DevQuasar/microsoft.Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | — | DevQuasar | 1 | 2.1k |
| tensorblock/microsoft_Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | tensorblock | 0 | 2k |
| AbteeXAILab/lumynax-reasoning-phi4-mini-gguf | quantized | gguf | — | — | mit | AbteeXAILab | 0 | 1.2k |
| mlx-community/Phi-4-mini-reasoning-bf16 | finetune | -bf16 | 3.84B | M | mit | mlx-community | 2 | 881 |
| tatsuyaaaaaaa/Phi-4-mini-reasoning-gguf | quantized | gguf | — | — | mit | tatsuyaaaaaaa | 0 | 947 |
| Pinkstack/Phi-4-mini-6b-merge | merge | — | 6.25B | M | mit | Pinkstack | 6 | 566 |
| raj-hitpromo/hit-promo-reasoner | finetune | — | 3.84B | M | mit | raj-hitpromo | 3 | 696 |
| heretic-org/Phi-4-mini-reasoning-heretic | finetune | — | 3.84B | M | mit | heretic-org | 5 | 490 |
| mlx-community/Phi-4-mini-reasoning-8bit | quantized | -8bit | 3.84B | M | mit | mlx-community | 0 | 669 |
| steadyflow/Phi-4-mini-reasoning-AWQ | quantized | AWQ | 3.84B | M | mit | steadyflow | 0 | 618 |
| NikolayKozloff/Phi-4-mini-reasoning-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | NikolayKozloff | 3 | 395 |
| mlx-community/Phi-4-mini-reasoning-6bit | quantized | — | 3.84B | M | mit | mlx-community | 1 | 413 |
| gopalparashar/phi4-mini-reasoning | quantized | GGUF | — | — | mit | gopalparashar | 0 | 443 |
| keisuke-miyako/Phi-4-mini-reasoning-gguf-q4_k_m | quantized | gguf | — | — | — | keisuke-miyako | 0 | 380 |
| vikrai/phi4-vik-rai | finetune | — | — | — | mit | vikrai | 1 | 326 |
| viniciusianni/Phi-4-mini-reasoning-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | viniciusianni | 0 | 360 |
| MCES10-Software/Phi-4-mini-reasoning-mlx-fp16 | finetune | mlx | 3.84B | M | mit | MCES10-Software | 0 | 351 |
| TitleOS/HomePhi4_4B_FP16 | finetune | — | 3.84B | M | mpl-2.0 | TitleOS | 1 | 296 |
| ldostadi/Phi-4-mini-reasoning-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | ldostadi | 0 | 318 |
| DuoNeural/Phi-4-Mini-Reasoning-Abliterated | finetune | — | 3.84B | M | mit | DuoNeural | 1 | 266 |
| SandLogicTechnologies/Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | SandLogicTechnologies | 1 | 248 |
| arcAI1/phi-4-mini-reasoning-mlx | finetune | mlx | 3.84B | M | mit | arcAI1 | 0 | 294 |
| Siarhei/Phi-4-mini-reasoning-4bit | quantized | -4bit | 3.84B | M | mit | Siarhei | 0 | 289 |
| Bifrost-AI/Phi-4-Mirage-sol-Reasoning-3.8B | finetune | — | 3.84B | M | mit | Bifrost-AI | 1 | 211 |
| viniciusianni/Phi-4-mini-reasoning-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | viniciusianni | 0 | 253 |
| amd/Phi-4-mini-reasoning-onnx-ryzenai-hybrid | quantized | — | — | — | mit | amd | 0 | 235 |
| JulianChang/Phi-4-mini-reasoning-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | JulianChang | 0 | 200 |
| jc-builds/Phi-4-mini-reasoning-GGUF | quantized | GGUF | — | — | mit | jc-builds | 0 | 192 |
| ianlav/Phi-4-mini-reasoning-openvino | finetune | — | — | — | mit | ianlav | 0 | 185 |
| Date | Commit | SHA |
|---|---|---|
| 2025-12-10 | Update README.md (#4) | 0e3b1e2 |
| 2025-12-10 | Upload data_summary_card.md (#3) | a7b58c7 |
| 2025-05-01 | Update urls | 7a8c4e2 |
| 2025-05-01 | update | ccda8a0 |
| 2025-05-01 | Update README.md | c4b3741 |
| 2025-04-30 | Upload 2 files | 9db0932 |
| 2025-04-29 | fixed inf example | 7196d37 |
| 2025-04-29 | added model files | 661e722 |
| 2025-04-29 | initial commit | 079d2f3 |
Hosted endpoints observed for Phi-4-mini-reasoning, 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-4-mini-reasoning | — | — | — | — | — | — | — | 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.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:40 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:11 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.
