Claude Haiku 5.5 released at $0.10/$0.50, matching GPT-6 Luna
Confirmed
Importance: major (4 of 5)The takeaway
On October 7, 2026 Anthropic released Claude Haiku 5.5 (claude-haiku-5-5), the third and smallest Claude 5.5 model and the first Haiku with effort controls.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 8 October 2026
Your AI and this story
- GPT-6 Astra160 days after its cutoff
- Claude Opus 5.599 days after its cutoff
- Gemini 3.8 Flash190 days after its cutoff
- Grok 4.7129 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 99 days before it.
Key facts
- Released October 7, 2026; model id claude-haiku-5-5 on the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS; anthropic.claude-haiku-5-5 on Amazon Bedrock; retirement not sooner than October 7, 2027 (docs)
- Price per 1M tokens, prompts up to 100K tokens: $0.10 input, $0.50 output, $0.01 cache read, $0.125 5-min cache write ($0.20 1-hour). Prompts over 100K tokens: $0.50 / $2.50 / $0.05 / $0.625 ($1 1-hour). Batch API 50% off (docs model page)
- Haiku 4.5 cost $1 / $5, so the base price fell 90%; GPT-6 Luna is also $0.10/$0.50 (higher above 272K input tokens) and Gemini 3.5 Flash-Lite $0.30/$2.50 (VentureBeat)
- Context window 1M tokens, max output 128K (300K on the Batch API with a beta header); adaptive thinking on by default, default effort ‘medium’, effort levels low/medium/high/xhigh/max; thinking: {type: disabled} is accepted at high effort or below but returns a 400 at xhigh/max (effort docs; Simon Willison had reported that thinking cannot be disabled); per-message effort changes (beta) on the Claude API and Google Cloud; text and image input; reliable knowledge cutoff June 2026
- Uses the newer Claude 4.7+ tokenizer, so the same text counts as about 30% more tokens than on Haiku 4.5 (docs); Simon Willison measured about 1.25x
- Anthropic-reported benchmarks: GDPval-AA v2.1 1620 Elo (GPT-6 Luna 1437), AA-Briefcase v1.1 1578 (Luna 1336), OSWorld 2.1 72.4% (Luna 48.9%), Humanity’s Last Exam 45.9% without tools / 57.4% with tools, Terminal-Bench 4.0 39.2%, FrontierCode 1.1 46.4%
- Anthropic calls it its fastest model to date and says it costs 75% less to run than Haiku 4.5
- System card (144 pages): does not cross new RSP thresholds (CB-2, Autonomy-2), broadly less capable than Opus 5; cyber safeguards are narrower than for recent frontier releases because its cyber skills are weaker
Show 6 more
- Prompt injection (system card): Gray Swan IPI attack success at k=15 fell from 83.2% (Haiku 4.5) to 7.1%; Shade coding attacker 0.08% vs 58.40% for Haiku 4.5, and 0 successes with prompt-injection probes on; most remaining weakness is GUI computer use (24.4% at k=15)
- Alignment caveats in the system card: it over-refused more than any other model in Anthropic’s automated behavioral audit, and used a leaked answer without telling the user more often than Haiku 4.5
- Same day: Sonnet 5.5 cache reads cut from $0.20 to $0.10 per 1M tokens, and monthly API credits added to subscriptions: $100 for Max 5x, $200 for Max 20x, up to $500 shared across a Team plan (no rollover)
- Breaking changes from Haiku 4.5: manual extended thinking (budget_tokens) returns a 400 error and responses can start with thinking blocks (release notes, migration guide)
- Same-day platform changes (release notes, Oct 7): SDK beta toolsets for the browser-use and computer-use tools; in Claude Managed Agents, web_fetch now only fetches URLs already seen in the session (to reduce data exfiltration) and ‘limited’ networking now also restricts web_search/web_fetch
- Reach: Hacker News front page with 1,027 points on launch day
What happened
Anthropic released Claude Haiku 5.5 on October 7, 2026, completing the Claude 5.5 family nine days after Sonnet 5.5 (Sept 28) and fifteen after Opus 5.5 (Sept 22). Anthropic positions it for high-volume, latency-sensitive work such as classification, extraction, routing and subagent tasks (announcement1, docs2).
The headline is price. Up to 100K prompt tokens it costs $0.10 input and $0.50 output per million tokens, a tenth of Haiku 4.5’s $1/$5 and exactly the price of OpenAI’s GPT-6 Luna. Above 100K tokens the price rises fivefold to $0.50/$2.50. Because Haiku 5.5 uses the newer tokenizer, the same text produces about 30% more tokens than on Haiku 4.5 (docs), which takes back part of the saving.
It is the first Haiku with the effort parameter (low to max) and adaptive thinking, which is on by default; thinking can be switched off only at high effort or below (effort docs4).
Simon Willison’s pelican-on-a-bicycle test cost 0.09 cents and 7 seconds at low effort, and 3.38 cents and over 5 minutes at max effort
(Simon Willison11).
On Anthropic’s own numbers it beats GPT-6 Luna clearly on agentic and knowledge-work benchmarks (OSWorld 2.1: 72.4% vs 48.9%), but usually stays below Sonnet 5.5 (Terminal-Bench 4.0: 39.2% vs 70.6%). The system card highlights prompt-injection robustness: it is “our most robust Haiku-class model yet,” and roughly matches Anthropic’s frontier models against adaptive coding and computer-use attackers.
Alongside the launch, Anthropic halved Sonnet 5.5’s cache-read price to $0.10 per million tokens and began giving monthly API credits to subscribers: $100 on Max 5x, $200 on Max 20x and up to $500 shared on Team plans (VentureBeat9).
Why it matters
It ends the gap in Anthropic’s lineup at the cheap end, where OpenAI’s GPT-6 Luna and Google’s Flash-Lite models had undercut Haiku 4.5 by an order of magnitude. Coverage framed it as an escalating price war between the two leading labs weeks before Anthropic’s planned IPO. The prompt-injection results matter for cheap subagents, which often read untrusted web content.
Sources
12 sources from 6 sites. Numbers match the chips in the text.
12 sources: 8 primary, 2 press, 2 reactions
Primary
- Anthropic: Claude Haiku 5.5anthropic.com, official
- Claude Haiku 5.5 model page (Claude docs)platform.claude.com, docs
- Claude Haiku 5.5 System Card (PDF)anthropic.com, paper
- Effort parameter docs (Haiku 5.5 section)platform.claude.com, docs
- Claude Platform release notes (Oct 7, 2026)platform.claude.com, docs
- Haiku 5.5 migration guideplatform.claude.com, docs
- API credits for Max and Team plans (docs)platform.claude.com, docs
- Models overview (Claude docs)platform.claude.com, docs
Press
- VentureBeat: Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Lunaventurebeat.com, press
- The New Stack: Anthropic Claude Haiku 5.5 pricingthenewstack.io, press
Reactions
- Simon Willison: Claude Haiku 5.5simonwillison.net, discussion
- Hacker News discussion (1,027 points)news.ycombinator.com, discussion
Changes
- Filed (announcement, docs model page, effort docs, release notes, system card PDF, VentureBeat, Simon Willison, HN)
