# MiniCPM5-2B ![Organization](https://img.shields.io/badge/Org-OpenBMB-555) ![Context](https://img.shields.io/badge/Context-131%2C072%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-Open%20Weights-2da44e) ![License](https://img.shields.io/badge/License-Apache%202.0-d29922) ![Modality](https://img.shields.io/badge/Modality-Text-8957e5) ![Released](https://img.shields.io/badge/Released-Sep%202026-555) > **Quick answer:** MiniCPM5-2B is OpenBMB's second model in the MiniCPM5 series — a dense 2B-parameter Transformer built for on-device and resource-constrained deployment. It reaches 2B-class open-source SOTA (average 53.9 across OpenBMB's comparison suite) and beats several larger 4B-class models on coding, math, tool use, and agentic tasks. Apache 2.0, 131K context, released September 2026. ## At a Glance **Where MiniCPM5-2B leads** - 2B-class open-source SOTA across coding, math, tool use, long-context, and agentic benchmarks per OpenBMB's comparison suite - Strong math reasoning for its size: 86.5% AIME 2026, 63.8% HMMT Feb 2026 - Native 131K context with reported long-context retrieval scores (59.0% AA-LCR) well above same-size peers **Where it lags** - SWE-bench Pro (14.4%) and Terminal-Bench v2.1 (8.6%) remain far below larger models — coding-agent tasks are still hard at this size - HLE score (8.9%) reflects the ceiling of a 2B model on frontier knowledge questions **Best for:** local/on-device assistants, coding agents, and tool-use workflows where a small footprint matters more than frontier-model ceiling performance. ## What MiniCPM5-2B Is MiniCPM5-2B is the second release in OpenBMB's MiniCPM5 family, following MiniCPM5-1B. It's a dense, standard LlamaForCausalLM architecture (2.52B total parameters, 42 layers, GQA with 16 query / 2 KV heads) trained with the same recipe as MiniCPM5-1B scaled up, plus a full "UltraData" tiered training pipeline (base → mid-train → SFT → RL → On-Policy Distillation). It ships alongside its full training data stack (UltraX pretraining corpus, UltraData-Code, UltraData-SFT-Agent-2609 with 500K agent samples, UltraData-RL-2609 with 80K+ RL samples) — one of the more transparent open releases in the small-model space. The RL + On-Policy Distillation stage is reported to add +10.96 points average on reasoning/general tasks and +6.96 on agentic tasks over the SFT-only checkpoint. ## Specifications | Field | Value | |---|---| | Organization | OpenBMB | | Parameters | 2.52B total (1.98B non-embedding), dense | | Context window | 131,072 tokens | | Architecture | Standard LlamaForCausalLM, 42 layers, GQA (16 Q / 2 KV heads) | | License | Apache 2.0 | | Release date | 2026-09-03 | | Knowledge cutoff | 2026-05 (approx.) | | Modality | Text | ## Pricing Open weights — free to download and self-host; cost is inference/hosting only. No hosted API pricing published by OpenBMB at release. ## Context Window MiniCPM5-2B has a **131,072-token context window** — roughly 250+ pages of text in a single request. OpenBMB reports 59.0% on AA-LCR (long-context retrieval), notably ahead of same-size peers (Qwen3.5-2B: 5.3%, Gemma-4-E2B-it: 28.7%), suggesting the long-context claim is backed by real retrieval performance rather than just an extended position-embedding ceiling. ## Public Benchmark Scores *Scores below are self-reported by OpenBMB in the MiniCPM5-2B model card, evaluated against a comparison set of same-size and adjacent-size open models. They are not Benchgen measurements. Figures marked † in the original card come from the official Artificial Analysis release; all others were reproduced internally by OpenBMB.* | Benchmark | Score | Source | Date | |---|---|---|---| | [AIME 2026](/benchmarks/mathematical-association-of-america/aime-2026) | 86.5 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [HMMT Feb 2026](/benchmarks/harvard-mit/hmmt-2026) | 63.8 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [GPQA Diamond](/benchmarks/rein-et-al/gpqa-diamond) | 70.2 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [Humanity's Last Exam](/benchmarks/centre-for-ai-safety/humanitys-last-exam) | 8.9 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [AA-LCR](/benchmarks/artificial-analysis/aa-lcr) | 59.0 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [BFCL v4](/benchmarks/uc-berkeley/bfcl-v4) | 66.6 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [SWE-bench Verified](/benchmarks/princeton-nlp-openai/swe-bench-verified) | 46.4 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [SWE-bench Pro](/benchmarks/scale-ai/swe-bench-pro) | 14.4 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [Terminal-Bench 2.1](/benchmarks/laude-institute/terminalbench) | 8.6 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [BrowseComp-ZH](/benchmarks/browsecomp-zh) | 43.5 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [GDPval-AA v2](/benchmarks/artificial-analysis/gdpval-aa-v2) | 19.6 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [LiveCodeBench v6](/benchmarks/livecodebench-v6) | 69.1 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [MMLU-Pro](/benchmarks/mmlu-pro) | 70.8 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [MMLU-Redux](/benchmarks/mmlu-redux) | 84.7 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [SuperGPQA](/benchmarks/supergpqa) | 40.8 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [IFEval](/benchmarks/ifeval) | 86.7 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [IFBench](/benchmarks/ifbench) | 66.3 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [Multi-IF](/benchmarks/multi-if) | 71.8 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [LongBench v2](/benchmarks/longbench-v2) | 43.7 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [MATH-500](/benchmarks/math-500) | 94.6 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [OJBench](/benchmarks/ojbench) | 32.5 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [τ³-Bench Banking](/benchmarks/tau3-banking) | 20.8 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | | [τ²-Bench Telecom](/benchmarks/tau2-telecom) | 97.1 | [OpenBMB model card](https://huggingface.co/openbmb/MiniCPM5-2B) | 2026-09 | **Not mapped (excluded, no matching Benchgen benchmark entity):** LCB-Pro 25Q2 (Easy/Medium), SciCode (wbg) — a variant reported alongside a different weighting than our SciCode entity, NoLiMa, LongBenchPro, GAIA Text-103, ClawGym-Bench, WildClaw, QwenClaw. These are excluded rather than force-mapped to a similarly-named existing entity, since methodology details (task counts, exact variant) were not independently confirmed to match. Flagged for a future dedicated benchmark-config pass if cross-model demand emerges. ## MiniCPM5-2B vs Alternatives | Model | Context | SWE-bench Verified | GPQA | Price (in/out per 1M) | |---|---|---|---|---| | **MiniCPM5-2B** | 131K | 46.4% | 70.2% | Open weights | | [MiniCPM5-1B](/models/openbmb/minicpm5-1b) | 131K | — | — | Open weights | | [Qwen3.5-2B](/models/qwen/qwen3-5-2b) | — | 6.0% | 55.8%† | Open weights | | [Gemma-4-E4B-it](/models/google/gemma-4-e4b) | — | 3.0% | 51.3%† | Open weights | *Rival scores are as reported in OpenBMB's own comparison table (same source as MiniCPM5-2B's scores above), not independently re-verified by Benchgen.* MiniCPM5-2B's pitch is that a well-executed dense 2B model with strong post-training (RL + OPD) can match or beat sloppier 4B-class releases on real capability signals like tool use and long-context retrieval — worth checking against your own workload before assuming bigger always wins on-device. ## How MiniCPM5-2B Performs on Real Agent Tasks MiniCPM5-2B's headline story is agentic capability at a small footprint: BFCL v4 (66.6%) and τ²-Bench Telecom (97.1%) both lead its comparison set, and the model's SFT data explicitly includes 500K agent-training samples (UltraData-SFT-Agent-2609). That said, coding-agent benchmarks that require sustained multi-turn execution (SWE-bench Pro 14.4%, Terminal-Bench 8.6%) show the ceiling is still well below larger models — tool-calling accuracy and long-horizon task completion are not the same capability, and builders evaluating this model for an agent harness should test both independently rather than extrapolating from one. ## Related Models & Benchmarks - Compare: [MiniCPM5-1B](/models/openbmb/minicpm5-1b) · [Qwen3.5-2B](/models/qwen/qwen3-5-2b) - Benchmarks: [GPQA Diamond](/benchmarks/rein-et-al/gpqa-diamond) · [AIME 2026](/benchmarks/mathematical-association-of-america/aime-2026) · [BFCL v4](/benchmarks/uc-berkeley/bfcl-v4) ## Frequently Asked Questions

What is MiniCPM5-2B?

MiniCPM5-2B is OpenBMB's dense 2B-parameter language model, the second release in the MiniCPM5 series, built for on-device and resource-constrained deployment with native 131K context.

What is MiniCPM5-2B's context window?

131,072 tokens (roughly 250+ pages of text), with reported long-context retrieval scores (59.0% AA-LCR) well ahead of same-size peers.

How much does MiniCPM5-2B cost?

It's open weights — free to download and self-host under Apache 2.0. There's no vendor-hosted API pricing at release.

Is MiniCPM5-2B open source?

Yes, released under the Apache 2.0 license, with training data (UltraX, UltraData-Code, UltraData-SFT-Agent-2609, UltraData-RL-2609) also open-sourced.

What is MiniCPM5-2B's knowledge cutoff?

Not explicitly disclosed by OpenBMB; approximate based on release timing (September 2026).

--- *Specs and scores sourced from OpenBMB's official MiniCPM5-2B model card. Third-party comparison scores attributed inline. Last updated 2026-09-08.*

MiniCPM5-2B — Evaluations

Evaluations