Daily drop
Checked claims for October 25, 2026
Top claim
VerifiedProvisionalrelease
Moonshot released Kimi K3.1 Pro, a 3T-class model with a 1M-token context window. It is on the Kimi API as kimi-k3.1-pro and kimi-k3.1-pro-highspeed, and in Kimi Code.
- How we checked
- Kimi's announcement blog confirms release, size class, context window, API model IDs and products
- Why it matters
- This is a new very large model you can call through an API today. The highspeed variant is listed as 2 to 3 times faster.
- Try this
- Run a few of your own coding tasks against kimi-k3.1-pro on the Kimi API Platform and compare cost and latency with your current model.
- Source record
- Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-11-08
Interconnects (Nathan Lambert)primary source
VerifiedProvisionalrelease
Gemini CLI published nightly pre-release v0.67.0-nightly.20261025 on Oct 25, 2026. It has no written release notes, only a compare link to the previous nightly.
How we checked · Why it matters
- How we checked
- GitHub release page shows the tag, date and pre-release status
- Why it matters
- This is a routine nightly build. The page does not list any new features.
- Source record
- Tier T1-seed · trust 0.5 (lower bound 0.196) · re-check 2026-11-08
Google DeepMind / Google AIprimary source
VerifiedProvisionalresearch
Artificial Analysis gives Kimi K3.1 Pro a score of 55 on its Intelligence Index, ranking it 4th of 215 models listed.
How we checked · Why it matters
- How we checked
- Read score and rank on the Artificial Analysis model page; models above it not named there
- Why it matters
- This is a third-party score, not the vendor's. The podcast notes that longer, exploratory benchmarks such as ProgramBench show weaker results.
- Source record
- Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-11-08
Interconnects (Nathan Lambert)independent source
VerifiedProvisionaltechnique
LiteLLM's root cause: a master key committed to a public repo on May 27, 2026 was force-pushed over but never revoked. An attacker used it on Oct 10 to read agent configs holding an org owner's GitHub token.
How we checked · Why it matters · Try this
- How we checked
- Read LiteLLM's attack-chain postmortem; dates and token path confirmed there
- Why it matters
- Rewriting git history does not kill a leaked key. Agent configs that store broad personal tokens become a path to your release pipeline.
- Try this
- Revoke any key that ever reached a public commit, even after a force-push. Give agents scoped service-account tokens instead of an owner's personal GitHub token.
- Source record
- Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-11-08
Interconnects (Nathan Lambert)primary source
Vendor claimProvisionalresearch
Moonshot says it removed cyber-related data from mid-training and RL for K3.1 Pro. It reports 81.0% pass@3 on its internal xdev-bench-hard, against 90.5% for a checkpoint without that removal.
How we checked · Why it matters
- How we checked
- Numbers and method read on Kimi's own blog; internal benchmark, not independently reproduced
- Why it matters
- A lab is cutting one capability out of an open-weight model on purpose. Expect weaker results on security-heavy agent tasks.
- Source record
- Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-11-08
Interconnects (Nathan Lambert)primary source
Today's list · All days · latest.md