# tool-eval-bench

A tool-calling quality benchmark for LLMs in agentic workflows, built for
self-hosted serving stacks: **vLLM**, **SGLang**, **LiteLLM**, **llama.cpp**, **Strata**,
**TabbyAPI**, **NInfer**, **TensorFold**, and hosted **Gemini** and **Anthropic**.

Each scenario observes one assistant conversation with mock tools. It does not
measure independent agents, delegation, or inter-agent handoffs. Localization
coverage is currently German-focused. Difficulty tiers are author estimates,
not calibrated model rankings.

It runs 69 deterministic scenarios (plus 23 opt-in Hard Mode ones) through
OpenAI-compatible `/v1/chat/completions` endpoints, scores each as pass, partial,
or fail, and writes a full conversation trace for every one. Throughput,
long-context retrieval, and accuracy benchmarks run against the same endpoint.

![tool-eval-bench benchmark output](docs/images/benchmark-output.png)

## Quickstart

### Install

```bash
uv tool install git+https://github.com/SeraphimSerapis/tool-eval-bench.git

# With throughput benchmarking (bundles llama-benchy)
uv tool install 'tool-eval-bench[perf] @ git+https://github.com/SeraphimSerapis/tool-eval-bench.git'
```

Also available via [Docker](docs/docker.md) if you would rather not have a local
Python, or as a [development checkout](CONTRIBUTING.md#development-setup).

### Run it

Point it at an OpenAI-compatible endpoint and run the core 15 scenarios. This
takes a couple of minutes and needs no configuration file:

```bash
tool-eval-bench run --short --base-url http://localhost:8000
```

Drop `--base-url` and it scans the common localhost ports used by vLLM,
llama.cpp, TensorFold, SGLang, LiteLLM, Ollama, and TGI. `tool-eval-bench probe` checks an
endpoint is reachable before you commit to a full run.

When that looks right, drop `--short` for the standard 69-scenario benchmark.
Pass `--seed` so the run is reproducible:

```bash
tool-eval-bench run --seed 42
```

TensorFold is detected from its model owner or metrics namespace and uses the
existing OpenAI-compatible adapter. Its default port is already scanned:

```bash
tool-eval-bench run --short --base-url http://127.0.0.1:8080/v1 --seed 42
```

See [TensorFold compatibility](docs/backends.md#tensorfold) for deployment
metadata limits, structured-output setup, and speculative metrics.

Engine detection does not infer identity from a port or generic health response.
A server's declared identity wins over compatibility shapes: model owner, `Server`
header, `/health` service, `/.well-known/serviceinfo`, or `/props` build name. That is
how Strata and TabbyAPI are told apart from the llama.cpp-style `/props` they both
serve. llama.cpp is recognized through its model owner, identifying header,
characteristic props/build fields, or metrics. Halogen Flash is recognized through its `halogen:`
metrics, even when it also exports llama.cpp-compatible names. Unidentified servers
use the `unknown` backend label rather than being called vLLM. The OpenAI-compatible
request format is unchanged; `--backend` still pins the reporting label.
The Python API runs the same detection when `backend` is left at `unknown`; an explicit
label is kept, and `probe_engine=False` skips detection.
See [backend identification](docs/backends.md#backend-identification).

### Exercise controlled fixture variants

```bash
tool-eval-bench run --hardmode --variant-seed 1 --seed 42
```

`--variant-seed` chooses versioned mock environments independently of the model's
sampling seed. Sixteen scenarios vary across all ten authoring packages; the
others remain controls. Variants include dry weather, delayed or failed jobs,
clarification followed by action, room capacities, transaction outcomes, pagination,
and changed identifiers. Private YAML packs remain unchanged. The selected fixture
metadata participates in `config_fingerprint`; resume rejects a different variant.

Reports include paired small/crowded toolset deltas when both scenarios ran.
TC-88 scores visible numeric constraints independently of reasoning-channel
availability, which appears in capability diagnostics. Structured-output diagnostics
say that schema enforcement was requested, without asserting the backend enforced it.
See [methodology](docs/methodology.md#controlled-fixture-variants) for coverage and limits.

### Read your report

Every completed run writes two artifacts, both relative to the directory you ran
from:

| Artifact | Path |
|---|---|
| Markdown report, with the full trace per scenario | `runs/YYYY/MM/<run_id>.md` |
| SQLite record, queryable | `data/benchmarks.sqlite` |

The terminal summary gives you the composite score, the star rating, and
per-category percentages. Three things are worth checking before you compare two
runs:

- **`completion_rate`.** Scenarios that measured the serving environment rather
  than the model are dropped from the score instead of counted as zero:
  timeouts, connection errors, a request the endpoint rejected before the model
  produced anything, known backend initialization failures such as llama.cpp
  sampler grammar errors, and scenarios needing a capability the endpoint does
  not have. TC-45 forces a tool call, so it is excluded on an endpoint that does
  not enforce `tool_choice="required"`, which costs one extra request per run
  to detect. A run graded on 60 of 69 scenarios is not comparable to one graded
  on all 69.
- **Safety warnings.** An observed unsafe action or disclosure is reported even
  outside the injection scenarios. Harmless incomplete work is not a safety
  violation. With a recorded violation and less than 50% in the safety group,
  the rating is capped at three stars.
- **`config_fingerprint`.** Runs group on the leaderboard only when their
  configuration and discovered deployment metadata match. The deployment
  metadata includes the engine version, context window, quantization, GPU count,
  server slot count, and speculative decoding mode. Two scores from different
  cohorts are not a comparison.

To read past runs back:

```bash
tool-eval-bench history          # recent runs
tool-eval-bench leaderboard      # ranked, grouped by comparable configuration
tool-eval-bench compare A B      # two persisted runs, side by side
```

### Point it at your server

For remote servers or non-standard ports, create a `.env` file:

```bash
TOOL_EVAL_BASE_URL=http://your-server:8080
# ...or host and port separately, used when BASE_URL is empty:
TOOL_EVAL_HOST=your-server
TOOL_EVAL_PORT=8080

TOOL_EVAL_MODEL=         # optional: auto-detected from /v1/models
TOOL_EVAL_API_KEY=       # optional
```

Priority order: CLI flags > environment variables > `.env` > auto-discovery.
Env vars set by a calling process are never overridden by a stale `.env`.

To keep several endpoints in one `.env` for A/B runs, scope them by name and
pick one with `--provider`:

```bash
TOOL_EVAL_GEMINI_BASE_URL=https://generativelanguage.googleapis.com
TOOL_EVAL_GEMINI_API_KEY=...
TOOL_EVAL_GEMINI_MODEL=gemini-2.5-pro

TOOL_EVAL_LOCAL_BASE_URL=http://gpu-box:8080
```

```bash
tool-eval-bench run --provider gemini
tool-eval-bench run --provider local
tool-eval-bench compare <run-a> <run-b>
```

The name is free-form. `gemini`, `openai`, and `anthropic` also set the
report's backend label. See `.env.example` for the vendor endpoints. An
Anthropic Messages endpoint (`api.anthropic.com`, or any URL ending in
`/messages`, such as OpenCode Zen's) is detected from the URL; `--format
anthropic` pins it for a gateway root that serves several formats.

## What it measures

| | What it tests | More |
|---|---|---|
| **Tool-call quality** | 69 scenarios across categories A–O: tool selection, parameter precision, multi-step chains, refusal, error recovery, localization, instruction following, safety and prompt injection, 52-tool namespaces, autonomous planning, structured output | [methodology](docs/methodology.md) |
| **Hard Mode** | 23 opt-in adversarial, stateful, and transactional scenarios for models that already score well, broken down by capability in reports | [hard-mode](docs/hard-mode.md) |
| **Throughput** | llama-bench-style prefill and generation speed, with depth and concurrency sweeps | [benchmarks](docs/benchmarks.md) |
| **Long-context retrieval** | Needle-in-a-haystack across a grid of context lengths and depths, reporting effective context | [needle](docs/needle.md) |
| **Context pressure** | Pre-fill a share of the window before each scenario to find where quality slips | [context-pressure](docs/context-pressure.md) |
| **Speculative decoding** | Acceptance rate, effective tokens per second, speedup, plus a live monitor | [speculative-decoding](docs/speculative-decoding.md) |
| **Accuracy** | GSM8K, MMLU, and IFEval through the same adapter | [benchmarks](docs/benchmarks.md) |
| **Decision models** | Single-pass option scoring on llama.cpp `/v1/systemone` against the Typed Decisions test split: accuracy, KL and Brier against soft gold distributions, and calibration, plus a live canary monitor | [decision-models](docs/decision-models.md) |

TC-74 accepts confirmation ranges such as `14:00–14:45` and `2:00–2:45 PM` for
its authorized 2pm, 45-minute event. Stated start and end times must match;
incorrect 12-hour times are rejected just like incorrect 24-hour times.

An asterisk on a prefill rate marks an estimate from time to first content token.
The [throughput guide](docs/benchmarks.md) explains when that estimate is used.

Mock tool responses carry realistic payload noise — extra metadata, timestamps,
nested objects — so a model has to extract the right field from a response
shaped like a real API's, not a hand-trimmed one.

> **Scope.** This measures *tool-calling quality*: whether a model picks the
> right tool, passes the right parameters, chains correctly, and respects error
> and safety boundaries. It is not a full agentic system benchmark. See
> [related work](docs/related-work.md) for how it compares to BFCL, PinchBench,
> and Claw-Eval.

### Scoring

Each scenario scores 2 (pass), 1 (partial), or 0 (fail). The final score is
`(points earned / max points) × 100`, so every scenario counts equally and
larger categories carry proportionally more weight.

| Score | Rating |
|---|---|
| 90–100 | ★★★★★ Excellent |
| 75–89 | ★★★★ Good |
| 60–74 | ★★★ Adequate |
| 40–59 | ★★ Weak |
| 0–39 | ★ Poor |

When a safety violation is recorded and the safety group scores below 50%, the
rating is capped at ★★★ regardless of the composite. `--weight-by-difficulty` computes an
alternative score that weights harder scenarios more heavily.

TC-62 accepts Acme revenue stated inline or in a revenue bullet immediately under
a clear Acme heading, including Markdown headings and blank-line spacing. It does
not carry that attribution across intervening text or another company heading.
Wrong amounts, percentages, negated claims, and quoted figures still do not qualify.

Full rationale, the category table, the difficulty tiers, and the evaluator
design: [docs/methodology.md](docs/methodology.md).

## Commands

| Command | Purpose |
|---|---|
| `run` | Run tool-call scenarios |
| `probe` | Check inference-server reachability |
| `bench` | Throughput, speculative-decoding, or context-pressure benchmarks |
| `plugin` | Run GSM8K, MMLU, IFEval, needle-in-a-haystack, or decision models |
| `spec-live` | Monitor speculative-decoding metrics |
| `decision-live` | Monitor a decision model with live canary probes |
| `compare` | Compare stored runs or Markdown reports |
| `history`, `leaderboard`, `export` | Inspect or export persisted results |
| `resume` | Continue an incomplete run |

```bash
# Smoke test — 5 scenarios
tool-eval-bench run --scenarios TC-01 TC-02 TC-03 TC-04 TC-05

# Full 92 — standard suite plus Hard Mode
tool-eval-bench run --seed 42 --hardmode

# Quality plus speed
tool-eval-bench bench --seed 42 --perf

# Statistical rigor — Pass@k / Pass^k across trials
tool-eval-bench bench --seed 42 --trials 3 --perf

# Long-context retrieval, chained onto a full sweep
tool-eval-bench --hardmode --seed 42 --perf --needle

# Safety and tool selection only, failing CI on a safety regression
tool-eval-bench run --categories K A --fail-on-safety

# Tag an execution so every report it generates is identifiable
tool-eval-bench run --label "nightly qwen3 2026-08" --trials 3
```

`tool-eval-bench COMMAND --help` lists a command's options; every flag and exit
code is in [docs/cli-reference.md](docs/cli-reference.md). Flat invocations
(`tool-eval-bench --short`, `--history`) remain supported.

Scenario IDs passed to `--scenarios` resolve against all 92, so `--scenarios
TC-85` works without `--hardmode` and takes precedence over `--short` and
`--categories`. Selection is validated before model discovery, so a typo fails
immediately rather than becoming an empty run.

Runs are checkpointed to SQLite as each scenario finishes, so a Ctrl-C costs you
only the scenario in flight — `tool-eval-bench resume RUN_ID` picks up the rest.
See [docs/artifacts.md](docs/artifacts.md).

`--system-prompt TEXT` (or `--system-prompt-file PATH`) replaces the built-in
"helpful assistant" system prompt for every scenario in a run — useful for
testing whether a stricter persona stops models from violating evaluator
contracts. The benchmark reference-date line is appended after the override and
stays authoritative, so relative-time scenarios keep working whatever the prompt
says about today. The override is recorded in the run config and its comparison
fingerprint — a run with a different prompt is a different cohort, and resuming
across a change is refused — and the run report marks that one was used.
[`examples/system-prompt-function.txt`](examples/system-prompt-function.txt) is a
minimal override that constrains the model to verified output.

### Optional answer audits

A separate decision model can audit what the model told the user, such as a
claim that a payment went through or a question asking which contact was meant:

```bash
tool-eval-bench run --hardmode --base-url http://localhost:8000/v1 \
  --decision-judge \
  --decision-judge-base-url http://localhost:8084/v1 \
  --decision-judge-model clef-flash
```

`--decision-judge` picks the set: `recommended` (11 scenarios, also the default
when only the connection flags are given) or `all` (17). The judge answers one
versioned question per audited scenario using `/v1/systemone`.
Official points, safety warnings, and ratings stay deterministic. SQLite and the
Markdown report keep the original result alongside the audit's probabilities,
exact input, versioned rubric, model, latency, and any disagreement. An audit
error is an unavailable judgment, not a model failure. Held-out scenarios are
never sent to the judge.

Terminal output names the judge when its request starts, then shows the finding,
probability, and agreement or disagreement for that scenario. Unaudited scenarios
and skipped requests get no placeholder. JSON mode emits `decision_audit_start`
and `decision_audit_result` progress events on stderr.

Set `TOOL_EVAL_DECISION_JUDGE_API_KEY` only if the judge requires authentication.
The benchmark model's credentials, headers, and reasoning are not forwarded.
Auditing is off unless both judge connection flags are supplied. Audited and
unaudited runs share a comparison fingerprint, since an audit never changes a
score. Probabilities are uncalibrated, and assistant text can steer the judge.
See [answer audit limits and API configuration](docs/decision-models.md#answer-audits).

## Programmatic API

```python
import asyncio
from tool_eval_bench.api import run_benchmark

result = asyncio.run(run_benchmark(
    model="Qwen/Qwen3-8B",
    base_url="http://localhost:8000",
    backend="vllm",
    short=True,           # core 15 scenarios
    persist=False,        # skip SQLite/Markdown (caller handles storage)
))

print(result["final_score"])   # e.g. 87
print(result["rating"])        # e.g. "★★★★ Good"
```

The call returns a versioned envelope with `final_score`, `rating`,
`safety_warnings`, `deployability`, and `total_scenarios`, alongside the full
per-scenario detail. For subprocess integration, `--json-file` writes results to
a file and emits JSONL progress events on stderr. Every parameter and returned
field: [docs/api.md](docs/api.md).

External tools can validate configuration against the published schema via
`tool_eval_bench.schema.get_schema()`.

Speculative-decoding detection and Prometheus counter helpers live in
`runner.spec_detection`, independently of the throughput and speculative
benchmark runners. Existing imports of these helpers from `runner.speculative`
remain supported.

## Documentation

- **Getting results:** [CLI reference](docs/cli-reference.md) ·
  [troubleshooting](docs/troubleshooting.md) ·
  [run IDs, artifacts, and labels](docs/artifacts.md)
- **The benchmarks:** [methodology](docs/methodology.md) ·
  [hard mode](docs/hard-mode.md) · [needle](docs/needle.md) ·
  [decision models](docs/decision-models.md) ·
  [context pressure](docs/context-pressure.md) ·
  [speculative decoding](docs/speculative-decoding.md) ·
  [accuracy and throughput](docs/benchmarks.md) ·
  [held-out packs](docs/scenario-packs.md)
- **Running it elsewhere:** [backends](docs/backends.md) · [Docker](docs/docker.md)
- **Building on it:** [Python API](docs/api.md) · [architecture](docs/architecture.md)
- **Contributing:** [CONTRIBUTING.md](CONTRIBUTING.md) ·
  [add a scenario](docs/adding-a-scenario.md) ·
  [add a plugin](docs/adding-a-plugin.md) ·
  [add a backend adapter](docs/adding-an-adapter.md)
- **Context:** [related work](docs/related-work.md) · [releasing](RELEASING.md)

## Contributing

A pull request runs lint, type checking, and the suite on Python 3.13 against
the committed `uv.lock`. Python 3.11 and Windows run after merge. The same gate,
locally:

```bash
.venv/bin/ruff check .
.venv/bin/ruff format --check .
.venv/bin/mypy
env -u FORCE_COLOR .venv/bin/python -m pytest tests/ \
  --ignore=tests/test_llama_benchy.py -m "not live" --randomly-seed=104729
```

Setup, the full quality bar, and the pull request checklist are in
[CONTRIBUTING.md](CONTRIBUTING.md). `CHANGELOG.md` is generated — record changes
as fragments under [`changelog.d/`](changelog.d/README.md).

## Credits

Scenario methodology adapted from
[ToolCall-15](https://github.com/stevibe/ToolCall-15) by
[stevibe](https://x.com/stevibe) (MIT License). Licensed under the
[MIT License](LICENSE).

### Third-party data

The decision-model benchmark ships the `test` split of
[Typed Decisions](https://huggingface.co/datasets/LocalLLaMA/typed-decisions),
published by the LocalLLaMA organization on Hugging Face under the Apache License
2.0, pinned to revision `e039ebffcc280174dd354227424fb2b249f191de`. The rows are
unmodified apart from format conversion. The dataset's license and a notice
describing the changes sit next to the data in
[`src/tool_eval_bench/plugins/decision/vendor/typed_decisions/`](src/tool_eval_bench/plugins/decision/vendor/typed_decisions/NOTICE)
and ship in every build. This project's MIT license does not cover that data. See
[decision models](docs/decision-models.md#third-party-data-typed-decisions).