Global AI, Malaysia Angle / AI news for Malaysia
Anthropic launches Claude Haiku 5.5 with lower prices and adjustable effort settings
Anthropic says Haiku 5.5 is its fastest and cheapest small model, priced 90 per cent below Haiku 4.5 for prompts up to 100,000 tokens. Malaysian firms should test it on their own tasks before switching.
In brief
- Anthropic released Claude Haiku 5.5 on 7 October 2026. It calls the model the cheapest, fastest and most capable small model it has released, with the model ID claude-haiku-5-5.[1]
- Prices are $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, against $1.00 and $5.00 for Haiku 4.5. Anthropic puts the average saving at around 75 per cent.[1]
- Anthropic's own table puts Haiku 5.5 behind Sonnet 5.5 on every listed test. It is the first Haiku-class model with an adjustable effort setting.[1]
What Anthropic released
Anthropic released Claude Haiku 5.5 on 7 October 2026. The company says it is built for high-volume, cost-sensitive work such as summaries, database queries and classification, and that it is the fastest Anthropic model at standard speed. It is available on Amazon Web Services, Google Cloud and Microsoft Azure, as well as Anthropic's own platform.[1]
This is a supplier announcement. The performance figures come from Anthropic's own table, and we ran no independent test.[1]
What the price cut actually covers
Anthropic's price table lists Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Haiku 4.5 costs $1.00 and $5.00. Above 100,000 tokens, Haiku 5.5 rises to $0.50 and $2.50. Sonnet 5.5 is listed at $2.00 and $10.00. All prices are in US dollars, so a ringgit bill will also move with the exchange rate.[1]
The headline saving is smaller than the list-price drop. Anthropic says Haiku 5.5 costs around 75 per cent less to run on average. Its footnote says 90 per cent of Haiku 4.5 requests were in the shorter group, and that a new tokenizer uses slightly more tokens per task. Anthropic also added a monthly API credit for Claude Max and Team subscribers, but gives no amount.[1]
What the benchmark table shows
Anthropic compares Haiku 5.5 with Haiku 4.5, OpenAI's GPT-6 Luna and, for reference, Sonnet 5.5. On GDPval-AA v2.1, a test of agents on professional work across 44 occupations, Haiku 5.5 scores 1620. Haiku 4.5 scores 735, GPT-6 Luna 1437 and Sonnet 5.5 1840. On the OSWorld 2.1 offline subset, a computer-use test, Haiku 5.5 scores 72.4 per cent, against 15.7, 48.9 and 83.9 per cent.[1]
Gaps remain on harder coding work. Haiku 5.5 scores 39.2 per cent on Terminal-Bench 4.0, while Sonnet 5.5 scores 70.6 per cent. Haiku 5.5 sits below Sonnet 5.5 on every test in the table. Anthropic also quotes customers such as Asana, HubSpot and Box reporting gains in their own tests. These are vendor-supplied statements, not independent audits.[1]
Effort settings, safeguards and availability
Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, so users can choose whether to optimise for cost or for intelligence. On safety, Anthropic reports major improvements in almost all of its alignment evaluations compared with Haiku 4.5. Its cybersecurity safeguards still block penetration testing, so security teams should expect some refusals.[1]
AWS published its own post, Introducing Claude Haiku 5.5 on AWS, on 7 October. Anthropic's supported countries page lists Malaysia, and we read it on 12 October. That page covers access to Claude. It does not say where a Malaysian firm's data would be processed.[1][2][3]
Why Malaysia should care
Malaysia is on Anthropic's supported countries list, so Malaysian firms can use Claude. The announcement names no Malaysian customer, ringgit price or local data-residency term. For SMEs the relevance is cost, because Anthropic lists summaries, classification and live customer support among the jobs the model targets.
SMEs running chat or support tools billed per token
List prices fall 90 per cent for prompts up to 100,000 tokens, in US dollars.[1]
Practical move: Re-run your own sample messages on Haiku 5.5 and compare cost and accuracy.
Developers building agents for Malaysian clients
Anthropic positions Haiku 5.5 as a fast subagent beside Sonnet 5.5 and Opus 5.5.[1]
Practical move: Let Haiku handle repetitive steps, and record where it fails.
Finance leads who approve AI budgets
The 75 per cent average saving depends on prompt length and a new tokenizer.[1]
Practical move: Ask your supplier for an estimate based on your actual prompt sizes.
What Malaysians can do now
- Business owners: ask your AI vendor whether Haiku 5.5 is available and whether the ringgit price has changed.
- Developers: test Haiku 5.5 on real tasks from your own work and compare quality, speed and cost.
- Finance teams: record the dollar prices and the 100,000-token threshold, so you can check invoices against them.
What we still do not know
What we still do not know
- How Haiku 5.5 handles Malay, Chinese or mixed-language customer messages, since Anthropic's table lists no such test.
- What the monthly API credit for Claude Max and Team subscribers is worth.
- Whether independent testers will reproduce Anthropic's figures.
Sources
- 1.Introducing Claude Haiku 5.5 Anthropic, 7 October 2026
- 2.Introducing Claude Haiku 5.5 on AWS Amazon Web Services, 7 October 2026
- 3.Supported countries and regions Anthropic
- 4.File:Secretary of State Peter Kyle visits Anthropic in San Francisco. (54098999333).jpg Wikimedia Commons, 25 October 2024
- 5.File:Penang Malaysia Mini-market-corner-Lebuh-Gereja-and-Lebuh-King-01.jpg Wikimedia Commons, 6 January 2014
- 6.File:Market at Jalan Benteng, Kuala Lumpur.jpg Wikimedia Commons, 18 July 2025
