DeepSeek-V3.2 vs Step 3.7 Flash

DeepSeek-V3.2 60

The chat template earns 10, as the format exists only as Python code to run. The tool-calling format earns 10 because DeepSeek documents the DSML format in code but names no runtime parser, which only vLLM's own recipe does. Trained tool calling earns the full 20 for several calls in one block. Structured output earns nothing, as JSON output isn't documented. Reasoning control earns 10 for thinking or chat mode per request with reasoning_content returned. Context earns 10 for 128K trained tokens.

Step 3.7 Flash 85

Chat template 20, chat_template.jinja in the repository (5,723 characters). Tool-calling format 25, the card names the parser to use in two runtimes, step3p5 in vLLM (with auto tool choice) and in SGLang, and both ship Step parsers in a release. Trained tool calling 20, the card presents tool orchestration as a main use, with multi-turn results, and the template renders every call in a turn. Structured output 0, Nothing on the card about JSON or schema-constrained output for the weights. Reasoning control 10, the template has a reasoning_effort setting and keeps reasoning_content, and the card names the step3p5 reasoning parser for vLLM and SGLang, which returns the reasoning apart from the answer. Context 10, 256K tokens per the card, and 262,144 positions in config.json. 20 + 25 + 20 + 0 + 10 + 10 = 85.

DeepSeek-V3.2 60

Memory earns nothing, because the only build from DeepSeek or a runtime's library is the 689.5 GB FP8 release, which needs about 827 GB with the fifth added. Runtimes earn the full 25 for released model code in vLLM, SGLang, llama.cpp and Transformers, and for Ollama 0.40.14, whose llama.cpp pin (b11963) carries the deepseek32 architecture with no gate in Ollama's server code, though its library entry stays retired; nothing from LM Studio. Builds earn 10 for the single FP8 precision DeepSeek publishes. Terms earn the full 15 for MIT with nothing added, and download the full 10 since the repository isn't gated. Hosted earns nothing, since DeepSeek's own API moved to V4 on 24 April 2026 and Hugging Face maps only DeepInfra as live for it on 4 November (Novita and Featherless in error), where the line needs two providers. Fireworks' model page says Serverless Not Supported and lists it only for on-demand deployment on dedicated GPUs, which is renting the hardware rather than calling a served model.

Step 3.7 Flash 70

Memory 5, the smallest build at 4 bits is StepFun's IQ4_XS GGUF, 104.99 GB with the 3.97 GB projector, and 108.96 GB × 1.2 = 130.8 GB, over what the reference machine can give a model and within 256 GB. The IQ3 and Q3 builds are below 4 bits and aren't read. Runtimes 20, Released vLLM 0.33.0, SGLang 0.5.23 and Transformers 5.22.2 have the step3p7 model, and llama.cpp b12242 has the architecture, with StepFun's own GGUF. Four at 5 each, and the card documents all four. Builds 15, StepFun publishes GGUF at five quantised precisions, an FP8 build and an NVFP4 build. Terms 15, Apache 2.0 in the card's metadata and its licence section, an OSI licence with nothing added. Download 10, the Hub API reports gated false for all four repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, StepFun's own API serves it, per the card and our StepFun API listing, and Featherless and DeepInfra are live for it on Hugging Face. 5 + 20 + 15 + 15 + 10 + 5 = 70.

DeepSeek-V3.2 50

File safety earns 15, safetensors only but 10 off because prompts are built by running DeepSeek's Python from the repository. Provenance earns the full 15 for the deepseek-ai organisation and a SHA-256 on every shard. Safety evaluation earns nothing, since no results are published anywhere. Agent guidance earns nothing, since the notes about the parser and the developer role are usage notes, read alike for all three DeepSeek listings. Licence risk keeps the full 15 for plain MIT. The security programme earns 5 for the security@deepseek.com reporting address on DeepSeek's site, with no public advisories.

Step 3.7 Flash 45

File safety 15, safetensors and GGUF only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the stepfun-ai organisation, which the card's StepFun platform and model page belong to (10). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations section and gives no guidance on prompt injection, instruction priority, refusals or safeguards. Licence risk 15, Apache 2.0 has no use policy, threshold or change clause. Security programme 0, no way to report a flaw on the card, and our StepFun API listing found no security.txt, bug bounty or disclosure policy. 15 + 15 + 0 + 0 + 15 + 0 = 45.

DeepSeek-V3.2 34

The model card earns 14 when read with DeepSeek's own V3.2 model card PDF, which adds intended use and a parameter count, leaving languages and a knowledge cut-off missing. Training data earns 5, a general description with no cut-off. Last change earns 5, as nothing in the repository has changed since 1 December 2025 (338 days on 4 November) and DeepSeek's deepseek-recipe has no V3.2 code. Responsiveness earns nothing, with no reply from a deepseek-ai member in any of the 40 most recent discussions. Version policy earns 5 for earlier versions kept apart without dated notes. File health earns 5 because loading and parsing reports wait without a reply.

Step 3.7 Flash 39

Model card 14, Intended use, the architecture, the parameters, the context and the language (in the metadata) are on the card. Training data 0, Nothing on the card about the training data. Last change 10, the last commit, on 3 June 2026, fixed the modelling code, 160 days before this check and within 180. Responsiveness 0, none of the 11 most recent discussions that aren't pull requests (29 May to 24 June 2026) has a reply from a stepfun-ai member, counted on 10 November by author flags and dates with a script, text not read. Version policy 5, Step-3.5-Flash stays in its own repositories and 3.7 came out in new ones, with no dated note on the older cards that we read. File health 10, Config, template and card agree on the architecture and the context, and June commits fixed the modelling code. 14 + 0 + 10 + 0 + 5 + 10 = 39.