Current outcome
Anthropic has launched Claude Haiku 5.5; independent evaluation shows a composite intelligence index score of 43, ahead of GPT-6 Luna's 38, but with higher token usage and cost at max effort; in market reaction, Zhipu and MiniMax shares dropped significantly on competitive pressure.
Progress timeline
3 material updates- #01
Anthropic发布Claude Haiku 5.5：价格降90%，多项评测超过GPT-6 Luna
Anthropic launched Claude Haiku 5.5, a lightweight agent-focused model that cuts API prices by 90% and claims benchmark wins over GPT-6 Luna in computer-use, professional-work, and coding evaluations.
Source evidence: theblockbeats
- #02
Haiku 5.5独立评测43分：遥遥领先GPT-6 Luna，但最高档Token消耗超3倍
Artificial Analysis released independent evaluation results for Claude Haiku 5.5: composite intelligence index score of 43, ahead of GPT-6 Luna's 38, but at maximum reasoning effort it outputs about 162k tokens per task (3.2x Luna) with estimated per-task cost around $0.21 vs Luna's $0.07; at high effort it scores 38 with token usage close to Luna.
State after update: Anthropic has launched Claude Haiku 5.5. Independent evaluation shows a composite intelligence index score of 43, ahead of GPT-6 Luna's 38, but at maximum reasoning effort token usage and cost are significantly higher; at high effort it performs near Luna with lower cost.
Source evidence: theblockbeats
- #03
Claude杀入国产低价模型腹地：智谱、MiniMax盘中大跌
Amid competitive pricing and performance pressure from the Claude Haiku 5.5 launch, Hong Kong-listed Zhipu and MiniMax shares fell more than 6% and 9% intraday.
State after update: Anthropic has launched Claude Haiku 5.5; independent evaluation shows a composite intelligence index score of 43, ahead of GPT-6 Luna's 38, but with higher token usage and cost at max effort; in market reaction, Zhipu and MiniMax shares dropped significantly on competitive pressure.
Source evidence: theblockbeats