Anthropic launches Claude Haiku 5.5, outperforms GPT‑6 Luna while matching its price

Anthropic’s New Haiku Model

Anthropic announced the release of Claude Haiku 5.5, the first update to the Haiku line in roughly a year.

The new model scores 43 on the Artificial Analysis Intelligence Index, a rise of 26 points from the previous Haiku release.

According to Anthropic, Haiku 5.5 introduces “effort settings” and adaptive thinking, features not present in Haiku 4.5.

Pre‑launch chatter on social media hinted at the upgrade, with @scaling01 posting “happy Haiku 5.5 day” before the official announcement.

Another user, @kimmonismus, noted the model’s appearance in a Claude Code update and predicted “Luna‑pricing,” later confirming the price when it went live.

Anthropic’s official account described the model as “the cheapest, fastest, and most capable small model we’ve ever released.”

In practice, Anthropic claims Haiku 5.5 runs about 75 % cheaper than Haiku 4.5 on average.

Pricing and Positioning

Haiku 5.5 is priced to align with OpenAI’s GPT‑6 Luna, making it a direct cost competitor.

Tiered pricing is applied based on token volume and cache reads, with under‑100K token usage costing $0.10 per input token and $0.50 per output token, and cache reads at $0.01 per million tokens.

For workloads exceeding 100K tokens, the rate rises to $0.50 per input token, $2.50 per output token, and $0.05 per million cache reads.

On the same day, Anthropic reduced Sonnet 5.5’s cache‑read fee from $0.20 to $0.10 per million tokens, a change Anthropic says translates to roughly a 20 % cost reduction for long‑running or agentic tasks.

Monthly Claude Platform API credits now come in three tiers: Max 5x at $100, Max 20x at $200, and a Team plan up to $500, all of which can be applied to any model, including Haiku 5.5.

Anthropic positions Haiku 5.5 as a sub‑agent to pair with Opus 5.5 or Sonnet 5.5 for high‑volume, cost‑sensitive operations such as summarisation, data compaction, and database queries.

Developer @mikeyk summed up the strategy: “Opus does the heavy thinking, Haiku does the high‑volume work.”

Ecosystem and Early Adoption

The model is immediately available on the Claude Platform and through Claude Code, allowing developers to integrate it into existing workflows.

Claude’s Python and TypeScript SDKs now embed computer‑use and browser‑use toolsets, handling the action loop and driver communication without custom code.

Partner integrations appeared on launch day, with Cursor.ai reporting “10x less than Haiku 4.5” on shorter requests and publishing comparative benchmarks.

GitHub Copilot in VS Code reportedly “matched Claude Sonnet 5 on many coding tasks while using fewer tokens and steps,” according to the @code account.

Devin, a side‑kick tool, achieved a 58.4 % score on FrontierCode 1.1, outperforming Sonnet 5 while costing roughly one‑eighth per task, and is recommended as a “sidekick” under an Opus 5.5 lead in Fusion.

Agent Arena added Haiku 5.5 to its Code, WebDev, Text, Document, and Vision categories, with performance scores pending.

OpenDocRouter also listed the model the same day, expanding its accessibility in document‑centric pipelines.

Independent evaluation by Artificial Analysis confirmed the Intelligence Index ranking, placing Haiku 5.5 slightly ahead of GLM‑5.3 Flash (42), Gemini 3.8 Flash (41), and GPT‑6 Luna (38), and comparable to Kimi K3.

Overall, the release signals Anthropic’s intent to compete directly with OpenAI in the small‑model segment while offering a more economical alternative for high‑throughput applications.

Why This Matters

Haiku 5.5 gives cost‑conscious developers a high‑performance, low‑price alternative to GPT‑6 Luna, potentially reshaping pricing dynamics in the small‑model market.