pragma.vision Technology observatory & strategic foresight

Verification register AI & Agents

Readiness verdict

Berkeley Function Calling Leaderboard V4 (BFCL V4)

A dated reading of what is claimed, reported, and independently verified in the current evidence.

As of
2026-08-20
Revision
1
Method
v1.3.0

Current reading

Readiness band and full integer triple

AI-assisted assembly · derived results

Readiness band

Mature

Primary summary from verified readiness

Confidence
75% · stale
Computed at
2026-08-20T10:54:06.022525+00:00
Claimed
90

Public ambition and stated capability

Reported
88

Observed practitioner reporting

Verified
83

Independently supported evidence

Gap
+7

Claimed leads verified

Evidence strength Strong

Decision

What the current evidence supports

Human editorial judgment · 2026-08-20

Adopt with guardrails

Why
It is the de-facto primary academic benchmark for function calling, actively maintained (updated April 2026) with realistic agentic categories — reliable as a comparative signal, provided we account for the ~75% ceiling and the FC-vs-prompt methodology split rather than treating scores as production readiness.
Next
Use BFCL V4 as one input when choosing a tool-calling model for agent work, but pin the commit/package version and read sub-category scores (multi-turn, agentic, format sensitivity) rather than just the headline number; pair with our own task-representative evals.

Constraints

Blockers

No named blocker is present in the current public projection.

Evidence summary

Derived counts

AI-assisted assembly

Total
12
Tier 1
0
Tier 2
4
Tier 3
8
Supports
4
Contradicts
2
Context
6
Latest observed
2026-06-01

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Revisions

Initial public reading

  1. 2026-07-19 Reading moved from mature to mature.

Your opinion

Tell us anything.

What works, what doesn't, what's missing — especially about our watches, lenses, and the register itself. Anonymous is fine; leave an email if you'd like a reply.