Sign in

Head to head · Open-weight models you run yourself · October 2026 research run

Agents-A1-4B vs Devstral Small 2

Agents-A1-4B and Devstral Small 2 score within a point of each other on agent readiness, 61 (C) and 61 (C). Devstral Small 2 leads on risk. Both do open-weight models you run yourself.

Best open-weight models you download and run yourself for AI agents · All 6359 open weights comparisons

Agents-A1-4BInternScience (Shanghai AI Laboratory) · Open-weight model C61/100#39 of 160 open-weight models Devstral Small 2Mistral AI · Open-weight model C61/100#60 of 160 open-weight models

Which one, for what

Agents-A1-4B C

Best for Local agent loops (search, research, tool calls) on a laptop, a small GPU or a phone-class machine, under a plain licence

Ahead on

  • Agent interface, 70 against 65

Watch for

The base isn't declared, and the technical report doesn't cover the 4B

Devstral Small 2 C

Best for Developers who want a dependable local coding agent on one consumer GPU or a 32 GB Mac

Ahead on

  • Risk, 60 against 55

Watch for

No limitations, safety results, languages or data on the card

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Risk Devstral Small 2 +5

Agents-A1-4B 55

File safety 25, safetensors and GGUF only and nothing from the repository runs. Provenance 15, the files are in the InternScience organisation its GitHub repository links to, with a SHA-256 for every weight file. Safety evaluation 0 and agent guidance 0, nothing published. Licence risk 15, Apache 2.0 through the chain from Qwen3.5-4B with no conditions added. Security programme 0, no policy, contact or security.txt.

Devstral Small 2 60

File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, files in Mistral's verified mistralai organisation, which Mistral's site links to per our Mistral Large 4 listing (10). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations or safety section, and nothing on prompt injection, instruction priority or what the model refuses. Licence risk 15, Apache 2.0 has no use policy, threshold or change clause. Security programme 5, mistral.ai's security.txt gives a security contact and a submission policy, as read for the Mistral Large 4 listing (5). 25 + 15 + 0 + 0 + 15 + 5 = 60.

Agent interface Agents-A1-4B +5

Agents-A1-4B 70

Chat template 20, a Jinja file in the repository. Tool-calling format 10, the card names the qwen3_coder parser but the format is only in the template. Trained tool calling 20, the card says it was trained for tool use and the template renders several calls in one turn. Structured output 0. Reasoning control 10, the template turns thinking off per request and the card's reasoning parser returns it apart from the answer. Context 10, 262,144 tokens.

Devstral Small 2 65

Chat template 20, chat_template.jinja in the repository (11,362 characters), with the tokenizer's tekken file for mistral-common. Tool-calling format 25, the card documents tools in the OpenAI function format with an example, names the mistral tool-call parser with auto tool choice for vLLM, shows the same for SGLang, and ships test prompts with the expected tool-call output. Trained tool calling 10, the card says the model excels at using tools, and its tests cover one tool call and calls made one at a time in sequence. Structured output 0, Nothing on the card about JSON or schema-constrained answers for this model. Reasoning control 0, no reasoning mode is documented, and a June 2026 commit removed stray thinking tokens from the files. Context 10, 256K tokens per the card, served at 262,144 in its vLLM command. 20 + 25 + 10 + 0 + 0 + 10 = 65.

Score by category

CategoryWeight this runAgents-A1-4BDevstral Small 2Edge
Capability35%pendingpendingpendingnot scored in this run
Reach25%259090even
Risk20%205560Devstral Small 2 +5
Agent interface10%107065Agents-A1-4B +5
Documentation and upkeep10%104346Devstral Small 2 +3
Negative events≤1500
TotalC 61/100C 61/100

Facts side by side

FactAgents-A1-4BDevstral Small 2
KindOpen-weight modelOpen-weight model
VendorInternScience (Shanghai AI Laboratory)Mistral AI
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceApache-2.0Apache-2.0
Read-only variant documentednono
llms.txtnono
Last release2026-07-142026-07-15
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Agent reviews3/5 (4)none

Verdicts

Agents-A1-4B

A small agent cousin for search, research and tool-calling loops, with vision and 262,144 tokens of context. It runs in about 4 GB from InternScience's Q4_K_M GGUF, in vLLM, SGLang, Transformers or llama.cpp. The main risks are an undeclared base, no safety results or agent guidance, and a template that adds its own dated system prompt when you send none.

Devstral Small 2

The easy one to run: a 24B coding agent under Apache 2.0 that five runtimes carry, 15 GB at 4 bits in Ollama, with a 256K context, image input and well-documented tool calling. It's eleven months old, calls tools one at a time, has no reasoning mode, and its card says nothing on limits, safety or data.

Before you call either

Agents-A1-4B

  1. Quickest start is llama-server -hf InternScience/Agents-A1-4B-Q4_K_M-GGUF --jinja, about 3.4 GB with the vision projector
  2. With vLLM use --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder, as on the card, and add --language-model-only for text-only work
  3. Always send your own system message, or the template inserts InternScience's default prompt with the date fixed at 2026-07-14
  4. The card recommends temperature 0.85, top_p 0.95, top_k 20, min_p 0 and presence_penalty 1.1, and there is no generation_config.json to set them
  5. Thinking is on by default, send chat_template_kwargs with enable_thinking false for short tasks, a switch the card doesn't mention
  6. The config declares a multi-token prediction layer that isn't shipped, so don't turn on MTP speculative decoding
  7. Validate JSON output yourself, since structured output isn't documented

Devstral Small 2

  1. ollama run devstral-small-2 is the quickest start, at 15 GB
  2. On vLLM pass --tool-call-parser mistral --enable-auto-tool-choice
  3. Send the system prompt from CHAT_SYSTEM_PROMPT.txt in the repository. The card says results depend on it
  4. Use a low temperature. The repository's default is 0.15
  5. Plan for one tool call per turn
  6. The same model is on our Mistral API listing if you'd rather not host it

Questions

Which is better for AI agents, Agents-A1-4B or Devstral Small 2?

Agents-A1-4B and Devstral Small 2 score within a point of each other on agent readiness, 61 (C) and 61 (C). Devstral Small 2 leads on risk.

Other comparisons with Agents-A1-4B or Devstral Small 2

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.

Get an auditSee a sample report