# Fara1.5-27B vs Mage-VL-Cowork

> Fara1.5-27B and Mage-VL-Cowork score within a point of each other for desktop app control, 61 and 61 out of 100. Prices, MCP, x402, uptime and agent notes side by side.

- Canonical: https://www.anchorterminal.com/compare/fara1-5-27b-vs-mage-vl-cowork
- Markdown: https://www.anchorterminal.com/compare/fara1-5-27b-vs-mage-vl-cowork.md (~3,450 tokens)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-10

Fara1.5-27B and Mage-VL-Cowork score within a point of each other on agent readiness, 61 (C) and 61 (C). Mage-VL-Cowork leads on reach and agent interface. Both do desktop app control.

- Fara1.5-27B: grade C, 61/100, rank #62 of 160 open-weight models. Markdown https://www.anchorterminal.com/tools/fara1-5-27b.md · JSON https://www.anchorterminal.com/api/v1/tools/fara1-5-27b.json
- Mage-VL-Cowork: grade C, 61/100, rank #74 of 160 open-weight models. Markdown https://www.anchorterminal.com/tools/mage-vl-cowork.md · JSON https://www.anchorterminal.com/api/v1/tools/mage-vl-cowork.json
- Best desktop and mobile app automation for AI agents: https://www.anchorterminal.com/best/desktop-automation/index.md
- All 55 desktop comparisons: https://www.anchorterminal.com/compare/desktop-automation/index.md

## Which one, for what

### Fara1.5-27B (C)

Good for: Developers building a supervised web-browsing agent who want open weights and clear safety behaviour

Ahead on:
- Risk, 75 against 65
- Documentation and upkeep, 65 against 54

Watch for: Screenshots only, so a page can mislead or instruct it

### Mage-VL-Cowork (C)

Good for: Agents with one GPU and a sandboxed Linux desktop that want a small local model to carry out office and file tasks by GUI and shell

Ahead on:
- Reach, 60 against 55
- Agent interface, 85 against 80

Watch for: Loading runs Microsoft's Python with trust_remote_code, and the recommended runtime is a container in a personal Docker Hub namespace


## Why the edge

### Documentation and upkeep: Fara1.5-27B +11

- Fara1.5-27B (65): Model card 20, Intended use, out-of-scope use and known limitations, the architecture, the parameters, the context, the language (English only) and the data cut-off (20 March 2026) are all on the card. Training data 15, A data section that names the kinds of source and the mix, the scale, the cut-off (last acquisition 20 March 2026) and the filtering (a verifier keeps only successful trajectories). Last change 10, the last commit, on 27 July 2026, is 106 days before this check and within 180. Responsiveness 5, Of the 5 discussions that aren't pull requests (22 to 29 July 2026), 1 has a reply from a microsoft member, which is a few, counted on 10 November by author flags and dates with a script, text not read. Version policy 5, the card says future updates will ship as separately versioned models with their own cards, and the earlier Fara model stays in its own repository. File health 10, Config, template and card agree on the architecture and the 262,144-token context. 20 + 15 + 10 + 5 + 5 + 10 = 65.
- Mage-VL-Cowork (54): The card covers intended use, architecture, parameters, context and language, and misses limitations and a cut-off (14). Training data is described with named sets and sizes but no cut-off (5). The files are four days old (20). Two questions opened on 31 October have no Microsoft reply (0). It's a first release (5), and the files agree with the card (10).

### Risk: Fara1.5-27B +10

- Fara1.5-27B (75): File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, the files sit in Microsoft's verified microsoft organisation, which the card's GitHub repository belongs to (10), and the Hub gives a SHA-256 for the large files (5). Safety evaluation 0, the card says the model was red-teamed on Azure for groundedness, jailbreak resistance, harmful content and copyright, and publishes no results. Agent guidance 10, the card says what the trained pauses and refusals cover (three kinds of critical point, with the actions it stops on, and refusal data for harmful tasks) and the safeguards to add: a sandbox with no host files or credentials, allow-listed browsing, live monitoring with a way to halt, and checking its output. Licence risk 15, MIT has no use policy, threshold or change clause. Security programme 10, the microsoft/fara repository has a SECURITY.md, Microsoft's standard route to MSRC (5), and MSRC publishes advisories in public, as recorded on our other Microsoft listings (5). 25 + 15 + 0 + 10 + 15 + 10 = 75.
- Mage-VL-Cowork (65): File safety gets 15, safetensors only but loaded with trust_remote_code. Provenance gets 15 for files in Microsoft's organisation with a hash for every file. Licence risk keeps 15 under Apache 2.0. The security programme gets 10 for the repository's SECURITY.md route to MSRC and MSRC's public advisories. Agent guidance gets 10 for a statement that it runs arbitrary commands and a list of safeguards (sandbox, human confirmation, no credentials, allow-listing against injection), with nothing on trained refusals. Safety evaluation scores 0, as none is published.

### Reach: Mage-VL-Cowork +5

- Fara1.5-27B (55): Memory 10, the only build from Microsoft is the BF16 safetensors, 54.71 GB, and 54.71 GB × 1.2 = 65.7 GB, which fits the reference machine but not 64 GB. bartowski's GGUF isn't read for this line. Runtimes 15, the architecture is Qwen3_5ForConditionalGeneration, which vLLM 0.33.0, SGLang 0.5.23 and Transformers 5.22.2 run from the safetensors files, three at 5 each. Builds 0, Microsoft publishes the BF16 weights only. Terms 15, MIT in the card's metadata and its licence section, an OSI licence with nothing added. Download 10, the Hub API reports gated false on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, the card links an Azure Foundry page for this model, Microsoft's own platform, and Featherless is live for it on Hugging Face. 10 + 15 + 0 + 15 + 10 + 5 = 55.
- Mage-VL-Cowork (60): Memory earns the full 30, since the only build is 9.50 GB of BF16 files that need about 11.4 GB with the fifth added. Runtimes get 5 for vLLM, whose released code (0.33.0) carries the LLaVA-OneVision-2 architecture this model's config names, though Microsoft doesn't document it. Transformers runs it only through Microsoft's remote code, SGLang only in Microsoft's own container, and llama.cpp has nothing for it. Builds score 0, with no quantised build from Microsoft. Terms get 15 for Apache 2.0 with nothing added, the base models' licences included. Download gets 10 for an ungated repository, and hosted too 0, with no inference provider mapped.

### Agent interface: Mage-VL-Cowork +5

- Fara1.5-27B (80): Chat template 20, chat_template.jinja in the repository (7,756 characters), Qwen 3.5's template. Tool-calling format 25, Microsoft documents the format: a tool_call block holding a JSON call to computer_use, a table of the supported actions, the system prompt to send, a serve command for vLLM and a worked loop, with the full schema in its harness. Trained tool calling 10, one action per step: the model sees a screenshot, emits one tool call, and waits for the next screenshot, the one-at-a-time band. Structured output 15, Read for a computer-use model as typed outputs, as the other one we list was read. Reasoning control 0, the card says each reply opens with a chain-of-thought block and documents no way to turn it off or budget it. Context 10, 262,144 tokens per the card's details table, and config.json gives the language model 262,144 positions. 20 + 25 + 10 + 15 + 0 + 10 = 80.
- Mage-VL-Cowork (85): Read for a computer-use model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. Microsoft gives the system prompt and both tool schemas verbatim, what comes back for each channel, the SGLang parser and its own parsing code (25). It was trained to emit up to 15 calls of one kind per reply (20). Each action is JSON against a documented schema (15). The chat template is in the repository (20). Reasoning control scores 0, as turning thinking off isn't documented, and context 5 for a 32,768-token trained window.

## Score by category

| Category | Weight | Fara1.5-27B | Mage-VL-Cowork | Edge |
| --- | --- | --- | --- | --- |
| Capability | 35%, pending | pending | pending | not scored in this run |
| Reach | 25% (25 this run) | 55 | 60 | Mage-VL-Cowork +5 |
| Risk | 20% (20 this run) | 75 | 65 | Fara1.5-27B +10 |
| Agent interface | 10% (10 this run) | 80 | 85 | Mage-VL-Cowork +5 |
| Documentation and upkeep | 10% (10 this run) | 65 | 54 | Fara1.5-27B +11 |
| Negative events | ≤15 | 0 | 0 | |
| **Total** | | **61 · C** | **61 · C** | |

## Facts side by side

| Fact | Fara1.5-27B | Mage-VL-Cowork |
| --- | --- | --- |
| Kind | Open-weight model | Open-weight model |
| Vendor | Microsoft | Microsoft |
| Hosted endpoint | no (local only) | no (local only) |
| Transports |  |  |
| Auth | None | None |
| Pricing | Free | Free |
| x402 | no | no |
| Licence | MIT | Apache 2.0 |
| Read-only variant documented | no | no |
| llms.txt | no | no |
| Last release | 2026-07-27 | 2026-10-30 |
| Terms last updated | no document linked | no document linked |
| Privacy policy last updated | no document linked | no document linked |

## Verdicts

**Fara1.5-27B.** A browser agent model with unusually careful documentation: what it does, what it's trained to stop at, what it gets wrong and how to box it in, under MIT. Treat it as a component for Microsoft's harness, in a sandbox, with a person on the irreversible steps. Safety testing is described but not published, and it's English only.

**Mage-VL-Cowork.** A small computer-use cousin that looks at a Linux desktop and answers each turn with clicks and keys or a shell command, under Apache 2.0, in about 12 GB at BF16. Microsoft documents only its SGLang container and Transformers. The risks are an example harness that runs shell commands unapproved, and a headline score that is Microsoft's own.

## Before you call either

### Fara1.5-27B

1. Run it inside Microsoft's harness (the fara CLI or MagenticLite), in a sandboxed browser with an allow-list. The card says not to give it open access on a machine with anything sensitive
2. Send the card's system prompt word for word. The model was trained against it
3. Match a 1440×900 screen for reliable clicks
4. Expect it to stop and ask before purchases, sign-ins, messages or personal data unless the user authorised them up front. Pass the question to the user; don't answer for them
5. Treat page text as untrusted. The model can be redirected by instructions on a page
6. It emits one action per turn at temperature 0, and ends with `terminate`

### Mage-VL-Cowork

1. Run it only in a disposable VM or container with no host mounts, credentials or local network, as the card says; OSWorld's Docker provider is the tested setting
2. Quickest local load, `pip install "transformers>=4.57.3" accelerate pillow torch torchvision qwen-vl-utils`, then AutoProcessor and AutoModelForCausalLM from microsoft/Mage-VL-Cowork with trust_remote_code=True and torch_dtype bfloat16
3. To serve it, download the repository and run `MODEL_DIR=/path/to/Mage-VL-Cowork bash osworld_eval/scripts/serve_sglang_docker.sh`, which pulls xiaoyizhang/mage-vl-cowork:0.1.0 and listens on 127.0.0.1:30001; the card's bare sglang.launch_server command binds 0.0.0.0
4. Use the card's MAGE_HYBRID_SYSTEM_PROMPT unchanged, render with apply_chat_template(enable_thinking=True), send base64 JPEG screenshots to SGLang's native POST /generate, temperature 0.7, top_p 0.8, max_new_tokens 4096
5. Coordinates come back on a 0 to 999 grid; scale x by width/1000 and y by height/1000. Answer computer_use calls with a new screenshot and bash calls with text in tool_response, keeping three screenshots and folding older turns into one-line summaries
6. Put a human confirmation step in front of bash and irreversible GUI actions yourself; the bundled MAGEHybridAgent has none
7. vLLM 0.32.0 registers the LlavaOnevision2 architecture, but Microsoft doesn't document it; test it before relying on it

## Questions

### Which is better for AI agents, Fara1.5-27B or Mage-VL-Cowork?

Fara1.5-27B and Mage-VL-Cowork score within a point of each other on agent readiness, 61 (C) and 61 (C). Mage-VL-Cowork leads on reach and agent interface.


## For agents

- This comparison as JSON: https://www.anchorterminal.com/compare/fara1-5-27b-vs-mage-vl-cowork.json, and with the fewest tokens: https://www.anchorterminal.com/compare/fara1-5-27b-vs-mage-vl-cowork.min.md
- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {"a": "fara1-5-27b", "b": "mage-vl-cowork"}`. From a terminal: `anchor compare fara1-5-27b mage-vl-cowork`
- Each listing in full: https://www.anchorterminal.com/api/v1/tools/fara1-5-27b.json and https://www.anchorterminal.com/api/v1/tools/mage-vl-cowork.json

## Other comparisons with Fara1.5-27B or Mage-VL-Cowork

- [Fara1.5-27B vs Fara1.5-4B](https://www.anchorterminal.com/compare/fara1-5-27b-vs-fara1-5-4b.md)
- [Fara1.5-27B vs Fara1.5-9B](https://www.anchorterminal.com/compare/fara1-5-27b-vs-fara1-5-9b.md)
- [Fara1.5-4B vs Mage-VL-Cowork](https://www.anchorterminal.com/compare/fara1-5-4b-vs-mage-vl-cowork.md)
- [Fara1.5-9B vs Mage-VL-Cowork](https://www.anchorterminal.com/compare/fara1-5-9b-vs-mage-vl-cowork.md)

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp