Claude Haiku 5.5 vs GPT-6 Luna: which is better?

For prompts up to 100,000 tokens they cost the same, $0.10 input and $0.50 output per million tokens, and Anthropic's launch benchmarks put Haiku 5.5 ahead: Terminal-Bench 4.0 39.2% vs 16.4%, OSWorld 2.1 (offline subset) 72.4% vs 48.9%, FrontierCode 1.1 46.4% vs 42.4% and GDPval-AA 1620 vs 1437. These are vendor numbers. Above 100,000 tokens Haiku 5.5 rises fivefold to $0.50/$2.50, while GPT-6 Luna rises only to $0.20/$0.75 above 272,000 tokens, so Luna is the better deal for long-context work.

How much does Claude Haiku 5.5 cost?

Claude Haiku 5.5, released October 7, 2026, costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, with cache reads at $0.01 and 5-minute cache writes at $0.125. For prompts over 100,000 tokens it costs $0.50 input and $2.50 output. Batch processing halves that to $0.05/$0.25. Haiku 4.5 cost $1/$5, and Anthropic says Haiku 5.5 is about 75% cheaper to run on average; Simon Willison measured its new tokenizer using about 1.25 times as many tokens as Haiku 4.5 for the same text.

Is Gemini 3.8 Flash better than Claude Haiku 5.5?

Gemini 3.8 Flash is a larger, more expensive tier: $0.75 input and $3.75 output per million tokens at its introductory rate through December 31, 2026, doubling to $1.50/$7.50 from January 1, 2027, with a 1 million token context, 64K max output and native audio and video input. Choose it for multimodal input such as audio and video or for Google Cloud integration; choose Haiku 5.5 or GPT-6 Luna for cheap, high-volume text and agent subtasks, where they cost about one seventh as much per token.

What is Claude Haiku 5.5 best for?

Anthropic positions it for high-volume, latency-sensitive work: summaries, compaction, classification, extraction, routing, database queries, customer support, browser use and subagent work under Opus 5.5 or Sonnet 5.5. It is Anthropic's first Haiku with adjustable effort (default medium), has a 1M-token context and 128K max output, and runs on the Claude API, AWS, Google Cloud and Microsoft Azure as claude-haiku-5-5. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding.

Quick Answer

Claude Haiku 5.5 vs GPT-6 Luna vs Gemini 3.8 Flash (Oct 2026)

Published: October 8, 2026

The short answer

Claude Haiku 5.5 (October 7, 2026) is now the best small model for prompts under 100,000 tokens: same $0.10/$0.50 price as GPT-6 Luna, higher scores on Anthropic’s launch benchmarks. Above 100K tokens, GPT-6 Luna is much cheaper. Gemini 3.8 Flash sits a tier higher in price and is the pick for audio and video input. Prices are list USD per million tokens from vendor pricing pages, read October 8, 2026; see current API prices for every model.

The comparison

	Claude Haiku 5.5	GPT-6 Luna	Gemini 3.8 Flash
Released	Oct 7, 2026	Sep 22, 2026	Sep 2, 2026
Input / output (short prompts)	$0.10 / $0.50 (≤100K)	$0.10 / $0.50 (≤272K)	$0.75 / $3.75 (intro, through Dec 31, 2026)
Long-prompt price	$0.50 / $2.50 (>100K)	$0.20 / $0.75 (>272K)	Same rate; $1.50 / $7.50 from Jan 1, 2027
Cached input	$0.01 (≤100K)	$0.01	See Google pricing
Batch	50% off ($0.05 / $0.25)	50% off	50% off
Context / max output	1M / 128K	1.05M / 128K	1M / 64K
Reasoning	Adaptive, effort levels, default medium	Effort levels	Thinking levels
Input modalities	Text, image	Text, image	Text, image, audio, video

Benchmarks (Anthropic’s launch table)

Benchmark	Haiku 5.5	GPT-6 Luna	Haiku 4.5	Sonnet 5.5
Terminal-Bench 4.0	39.2%	16.4%	0.0%	70.6%
OSWorld 2.1 (offline subset)	72.4%	48.9%	15.7%	83.9%
FrontierCode 1.1 (Main)	46.4%	42.4%	—	52.1% (xhigh)
GDPval-AA v2.1	1620	1437	735	1840
Chartography (no tools)	46.4%	29.1%	6.4%	61.6%

These are vendor-reported. No independent Artificial Analysis Intelligence Index score for Haiku 5.5 had been published when we checked on October 8, 2026.

What changed with Haiku 5.5

Price. Input fell from $1 to $0.10 and output from $5 to $0.50 per million tokens for prompts up to 100K — 90% off list. Anthropic says prompts under 100K make up about 90% of requests to the previous Haiku and that Haiku 5.5 costs about 75% less to run on average.
A hidden cost. The new tokenizer counts more tokens for the same text; Simon Willison measured about 1.25x versus Haiku 4.5. Reasoning cannot be switched off and defaults to medium effort, so output tokens are higher than a non-reasoning model’s.
The 100K cliff. A 101K-token prompt pays five times the rate for the whole request. Keep RAG contexts and conversation history under the line, or route long prompts to Luna.
Alongside it, Anthropic halved Sonnet 5.5 cache reads to $0.10 per million (about 20% cheaper on agentic work) and is adding monthly API credits for Max ($100 or $200) and Team (up to $500) subscribers.

Which to pick

High-volume classification, extraction, summaries, support under 100K tokens: Haiku 5.5.
Long documents or long agent histories above 100K tokens: GPT-6 Luna.
Audio or video input, Google Cloud stack: Gemini 3.8 Flash — and budget for the January 2027 doubling.
Subagents under Opus 5.5 or Sonnet 5.5: Haiku 5.5, which Anthropic built for that role.

Test on your own prompts at the effort level you will run: per-task cost depends on tokens consumed, not just list price — see token efficiency vs token price. Earlier budget-tier comparison: GPT-6 Luna vs Gemini 3.8 Flash vs DeepSeek V4.1 Flash.

Last verified: October 8, 2026.

The short answer

The comparison

Benchmarks (Anthropic’s launch table)

What changed with Haiku 5.5

Which to pick

Sources