APIFlow-Bench 1.0

APIFlow-Bench 1.0 — first public release of the agentic API-workflow benchmark.

• Code + frozen task bank: the Inspect AI eval harness, the frozen 1.0 bank (467 tasks across 13 mock-API worlds), scoring, and the leaderboard site builder — see the README and CHANGELOG.
• Leaderboard (19-model panel, 5 epochs): https://www.postman.com/ai/apiflow-leaderboard/
• Raw trial transcripts (all 44,362 trials): https://github.com/postmanlabs/apiflow-bench-transcripts — per-model tar.gz bundles are attached to the transcripts-<model> releases in this repository.

Release provenance (bank content sha256, epoch count, panel configs) is pinned in registry.json.

Transcripts — qwen3p7-plus

Raw per-trial transcripts for qwen3p7-plus on APIFlow-Bench 1.0.

• Bulk download: transcripts-qwen3p7-plus.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/qwen3p7-plus/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — qwen3p6-plus

Raw per-trial transcripts for qwen3p6-plus on APIFlow-Bench 1.0.

• Bulk download: transcripts-qwen3p6-plus.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/qwen3p6-plus/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — minimax-m2p7

Raw per-trial transcripts for minimax-m2p7 on APIFlow-Bench 1.0.

• Bulk download: transcripts-minimax-m2p7.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/minimax-m2p7/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — kimi-k2p7-code

Raw per-trial transcripts for kimi-k2p7-code on APIFlow-Bench 1.0.

• Bulk download: transcripts-kimi-k2p7-code.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/kimi-k2p7-code/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — kimi-k2p6

Raw per-trial transcripts for kimi-k2p6 on APIFlow-Bench 1.0.

• Bulk download: transcripts-kimi-k2p6.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/kimi-k2p6/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — kimi-k2p5

Raw per-trial transcripts for kimi-k2p5 on APIFlow-Bench 1.0.

• Bulk download: transcripts-kimi-k2p5.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/kimi-k2p5/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — gpt-oss-20b

Raw per-trial transcripts for gpt-oss-20b on APIFlow-Bench 1.0.

• Bulk download: transcripts-gpt-oss-20b.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/gpt-oss-20b/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — gpt-oss-120b

Raw per-trial transcripts for gpt-oss-120b on APIFlow-Bench 1.0.

• Bulk download: transcripts-gpt-oss-120b.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/gpt-oss-120b/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.

Transcripts — gpt-5.6-terra

Raw per-trial transcripts for gpt-5.6-terra on APIFlow-Bench 1.0.

• Bulk download: transcripts-gpt-5.6-terra.tar.gz (attached) — every trial for this model.
• Individual files: served from the transcripts repo — https://github.com/postmanlabs/apiflow-bench-transcripts/raw/main/gpt-5.6-terra/<task_id>-t<N>.json (trial numbering matches the leaderboard's #trial-N).

Each JSON carries the full conversation (messages, reasoning blocks, tool calls and results), token usage, duration, and the verifier output as stored.