CrewAI 1.15.23 (2026-09-28) adds native Gemini 3.8 Flash support; crewai eval can score the last AMP-traced run (not just print it); task spans capture declared output format/results; LLM retries throttled providers and Bedrock acall falls back to sync. Tightens the observe→evaluate loop for multi-agent crews.
Key Takeaways
- ✓Release 1.15.23 / PyPI
- ✓Native Gemini 3.8 Flash via Google Gen AI SDK (LLM docs)
- ✓
crewai evalscores the last AMP-traced run; TUI Evaluate button + graded areas - ✓Task spans record declared output format and results
- ✓Retries on throttled providers; Bedrock
acall→sync fallback; SQLite/S3/Selenium cleanup fixes
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
Crew frameworks lag new Gemini IDs and often stop at logs—no tight last-trace→score loop. Throttling and Bedrock async/sync mismatches break long runs.
What shipped
1.15.23 adds native Gemini 3.8 Flash, AMP evaluation of the last traced run via crewai eval, richer task spans (output format + results), provider retry, and Bedrock acall→sync fallback, plus resource-cleanup fixes.
Benchmarks
No framework-only public scores in this patch; judge on your task suite and Google’s model card for 3.8 Flash.
Get started
pip install -U crewai==1.15.23, set Gemini via LLM docs, enable tracing, then crewai eval on the last AMP run. Repo: crewAIInc/crewAI.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.