Anthropic releases Claude Haiku 5.5, says running costs about 75% below Haiku 4.5
In short
Anthropic released Claude Haiku 5.5 on October 7, a small model the company says costs about 75% less to run on average than Haiku 4.5. For requests up to 100,000 tokens it is priced at $0.10 per million input tokens and $0.50 per million output tokens. Anthropic also halved the cache-read price of Sonnet 5.5 to $0.10 per million tokens.
Read the full story 2 min read
Anthropic released Claude Haiku 5.5 on October 7, IT Home and Jiemian reported. The company calls it the fastest and cheapest Haiku model so far, and Jiemian said it is designed for high-frequency, cost-sensitive tasks. According to Anthropic, Haiku 5.5 costs about 75% less to run on average than Haiku 4.5. [ 1 , 2 ]
For requests up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Above that threshold the prices rise to $0.50 and $2.50. TMTPost put the cache-read price at $0.01, or $0.05 for long prompts. It said Haiku 4.5's prices were $1, $5 and $0.10, which makes the short-context cut 90%, while the cut for typical workloads is about 75%. [ 1 , 3 ]
Anthropic also halved the cache-read price of Sonnet 5.5, from $0.20 to $0.10 per million tokens. IT Home reported that the company says this lowers Sonnet 5.5's running cost on most agent tasks by about 20%. TMTPost also reported that from this week, Claude Max and Team subscribers get monthly Claude Platform credits: $100 for Max 5x, $200 for Max 20x and up to $500 shared for Team. [ 1 , 2 , 3 ]
Anthropic's published results put Haiku 5.5 at 1620 on GDPval-AA v2.1, 1578 on AA-Briefcase v1.1, 72.4% on OSWorld 2.1 and 39.2% on Terminal-Bench 4.0. IT Home said the OSWorld score was on an offline subset. TMTPost compared these with OpenAI's GPT-6 Luna, which it put at 48.9% on OSWorld 2.1 and 16.4% on Terminal-Bench 4.0. It added that Luna's short-context prices are also $0.10, $0.50 and $0.01. [ 1 , 3 ]
Haiku 5.5 is the first Haiku model with adjustable effort settings, and Anthropic positions it as a subagent for Opus 5.5 and Sonnet 5.5. TMTPost said the context window grew from 200,000 to 1 million tokens and maximum output from 64,000 to 128,000 tokens. It also reported that the model adds Browser Use and a new classifier-based refusal mechanism. According to TMTPost, this is the third 5.5 model in about half a month, after Opus 5.5 on September 22 and Sonnet 5.5 on September 28. [ 1 , 3 ]
Why it matters
Chinese business and tech outlets covered the release as a price cut aimed at high-volume, cost-sensitive AI workloads. TMTPost set it directly against OpenAI's GPT-6 Luna, which it said now has nearly identical prices, pointing to price competition among model makers for this tier.
Key facts
- Anthropic released Claude Haiku 5.5 on October 7. [ 1 , 2 ]
- Anthropic said Haiku 5.5 costs about 75% less to run on average than Haiku 4.5. [ 1 , 2 , 3 ]
- For requests up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens; above that, $0.50 and $2.50. [ 1 , 3 ]
- Anthropic cut the Sonnet 5.5 cache-read price from $0.20 to $0.10 per million tokens. [ 1 , 2 , 3 ]
- Haiku 5.5 scored 1620 on GDPval-AA v2.1, 1578 on AA-Briefcase v1.1, 72.4% on OSWorld 2.1 and 39.2% on Terminal-Bench 4.0, according to Anthropic's figures. [ 1 , 3 ]
- Haiku 5.5 is the first Haiku model with adjustable effort settings. [ 1 , 3 ]
- Anthropic positions Haiku 5.5 as a subagent model for Opus 5.5 and Sonnet 5.5. [ 1 , 3 ]
- TMTPost reported the context window rose from 200,000 tokens in Haiku 4.5 to 1 million, and maximum output from 64,000 to 128,000 tokens. [ 3 ]
Confirmed by several sources
- Anthropic released Claude Haiku 5.5 on October 7. [ 1 , 2 ]
- Anthropic says Haiku 5.5's running cost is on average about 75% lower than Haiku 4.5's. [ 1 , 2 , 3 ]
- Short-context pricing is $0.10 per million input tokens and $0.50 per million output tokens, rising to $0.50 and $2.50 above 100,000 tokens. [ 1 , 3 ]
- Sonnet 5.5's cache-read price was halved to $0.10 per million tokens. [ 1 , 2 , 3 ]
- Haiku 5.5 scored 72.4% on OSWorld 2.1 and 39.2% on Terminal-Bench 4.0 in Anthropic's tests. [ 1 , 3 ]
- Haiku 5.5 adds adjustable effort settings and is positioned as a subagent for Opus 5.5 and Sonnet 5.5. [ 1 , 3 ]
Still unclear
- How large the price cut is: about 75% on average, or 90% for short-context calls. Anthropic and all three outlets give about 75% for typical workloads; TMTPost alone says list prices for short-context calls fell 90%, from $1 and $5 per million input and output tokens for Haiku 4.5.
- Monthly Claude Platform credits for Max and Team subscribers ($100 for Max 5x, $200 for Max 20x, up to $500 shared for Team). Reported only by TMTPost.
- Haiku 5.5's advantage over OpenAI's GPT-6 Luna on benchmarks. Only TMTPost makes the comparison, and it rests on test results Anthropic published.
- Cache-read prices for Haiku 5.5 of $0.01 per million tokens, or $0.05 above 100,000 tokens. Reported only by TMTPost.