Daily drop

Checked claims for October 17, 2026

Top claim

VerifiedProvisionalrelease

Claude Haiku 5.5 API pricing is $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, and $0.50/$2.50 above that. Haiku 4.5 costs $1/$5.

How we checked: Anthropic pricing page lists both Haiku 5.5 tiers and the Haiku 4.5 price
Why it matters: For short prompts that is 90% cheaper than Haiku 4.5, which changes the cost of high-volume classification and sub-agent calls.
Try this: Check how long your typical Haiku prompts are against the 100K-token price break before switching.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

VerifiedProvisionalrelease

How we checked · Why it matters · Try this

How we checked: Gemini API pricing page; the 1M output-limit claim was not confirmed there and is left out
Why it matters: The price doubles after the intro period, so cost estimates for anything built now should use the December prices.
Try this: Use the December 1 prices of $4/$20 per million tokens when you budget Gemini 4 Argon workloads.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-12-15

VerifiedProvisionalrelease

How we checked · Why it matters · Try this

How we checked: Hugging Face model card confirms specs and license terms; benchmark numbers left out as vendor claims
Why it matters: A large US open-weight multimodal model, but its license terms are stricter than Apache or MIT.
Try this: Read the Muse Model License on the model card before you build a product on it.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

VerifiedProvisionalrelease

How we checked · Why it matters · Try this

How we checked: Hugging Face model card shows size, license and precisions; DFlash2 and Mirror CLI not confirmed there
Why it matters: A large open model under a permissive license, which is simpler for commercial use than custom model licenses.
Try this: Compare the Apache 2.0 license on Beam's model card with the license of the open model you currently use.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

VerifiedProvisionalrelease

How we checked · Why it matters · Try this

How we checked: Hugging Face model card confirms size, context and license; the leak claim from Reddit commenters is unverified and left out
Why it matters: Earlier StepFun models were Apache 2.0, so check the new license terms before using this one commercially.
Try this: Read the StepFun Model License 1.0 terms on the model card before commercial use.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

VerifiedProvisionaltechnique

How we checked · Why it matters · Try this

How we checked: Read the original claude.dev post (Oct 15, 2026); authors and review flow match the quote
Why it matters: It is a written, detailed review pipeline for teams where agents write most of the code, including when a human still has to approve.
Try this: Read the post's section on risk-based approval, then write down which paths in your repo should always need a human reviewer.
Source record: Tier T1-seed · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

Docs-verifiedProvisionaltool

How we checked · Why it matters · Try this

How we checked: Found the changelog entry under version 2.1.300; not tested hands-on
Why it matters: Smaller request bodies can cut latency on long sessions. The setting matters if a proxy or gateway rejects compressed bodies.
Try this: If you route Claude Code through a proxy and requests start failing after upgrading, try CLAUDE_CODE_GZIP_REQUEST_BODIES=0.
Source record: Tier T3 · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

Docs-verifiedProvisionaltool

How we checked · Why it matters · Try this

How we checked: Merged PR page and nightly release notes; exact banner wording not visible, Antigravity pointer only in bot summary
Why it matters: If you use Gemini 4 Argon from the terminal, Google itself says Gemini CLI may not be the right tool.
Try this: If you plan to use Gemini 4 Argon from the terminal, read the updated Gemini CLI README first.
Source record: Tier T1-seed · trust 0.5 (lower bound 0.196) · re-check 2026-10-31

Docs-verifiedProvisionaltool

How we checked · Why it matters

How we checked: GitHub release notes in snapshot list PRs #29773, #29790, #29757; not tested
Why it matters: These are small reliability fixes for people who script or customize Gemini CLI. They are nightly builds, not stable.
Source record: Tier T1-seed · trust 0.5 (lower bound 0.196) · re-check 2026-10-31