Haiku 5.5 is real, which makes yesterday's "beats Opus" screenshots fiction. Anthropic itself says Sonnet still leads on the benchmarks it chose to show.
The headline is $0.10/$0.50 per million tokens, a price GPT-6 Luna matches to the cent. But look at the new effort dial: Anthropic's Terminal-Bench chart reportedly runs about 20% at medium and 39% at max.
Price per token is now the vanity metric. The bill is price per solved task, and the effort slider is where it hides.
https://venturebeat.com/technology/anthropic-launches-claude-haiku-5-5-with-90-api-price-reduction-matching-gpt-6-luna
#AI #LLM
Harvey Specter / Harvey
npub10q…u0dn0
2026-10-07 22:57:21 UTC
Author Public Key
npub10q864fp90wfjnkac8k226myaf5ewpxmz5e0d66lm27kq3j22928qcu0dn0Seen on
wss://nos.lolPublished at
2026-10-07 22:57:21 UTCEvent JSON
{
"id": "ba5a4a051f7bd7fbf5559a417d859c6ddfe113eb4626717aaebee5ec2feee40d",
"pubkey": "780faaa4257b9329dbb83d94ad6c9d4d32e09b62a65edd6bfb57ac08c94a2a8e",
"created_at": 1791413841,
"kind": 1,
"tags": [
[
"t",
"ai"
],
[
"t",
"llm"
],
[
"r",
"https://venturebeat.com/technology/anthropic-launches-claude-haiku-5-5-with-90-api-price-reduction-matching-gpt-6-luna"
]
],
"content": "Haiku 5.5 is real, which makes yesterday's \"beats Opus\" screenshots fiction. Anthropic itself says Sonnet still leads on the benchmarks it chose to show.\n\nThe headline is $0.10/$0.50 per million tokens, a price GPT-6 Luna matches to the cent. But look at the new effort dial: Anthropic's Terminal-Bench chart reportedly runs about 20% at medium and 39% at max.\n\nPrice per token is now the vanity metric. The bill is price per solved task, and the effort slider is where it hides.\n\nhttps://venturebeat.com/technology/anthropic-launches-claude-haiku-5-5-with-90-api-price-reduction-matching-gpt-6-luna\n\n#AI #LLM",
"sig": "fc31a19be4cc0bb4357c8da441a4402fd001163aee5b1c0f12ed3a3ebd461b94e7ed58309791d13fade3ad506e36aee10c6bb571ecdd9ae47d635ade43702ea2"
}