Claude Haiku 5.5 vs GPT-6 Luna

Claude Haiku 5.5 and GPT-6 Luna share a list price: $0.10/$0.50 per million input/output tokens for Haiku prompts up to 100K tokens, with $0.01 cache reads and $0.125 cache writes on both. The difference is where that price ends. Haiku 5.5 moves to $0.50/$2.50 over 100K tokens; Luna holds its base rate until 272K input tokens, then charges 2x input and 1.5x output. On Anthropic’s launch table, Haiku 5.5 scores higher on every shared row. On Anthropic’s own cost charts, Luna is cheaper per task at every GDPval-AA effort level and at four of five on OSWorld; at OSWorld medium, Haiku 5.5 is marginally cheaper.

Try Apidog today

This guide compares price, specs, shared benchmarks, per-effort cost curves, and workload routing. For individual model details, see what is Claude Haiku 5.5 and what is GPT-6 Luna. The implementation section sends the same prompt to both models in Apidog.

Price side by side

Per 1M tokens	Claude Haiku 5.5	GPT-6 Luna
Input, base tier	$0.10 (prompts up to 100K tokens)	$0.10 (up to 272K input tokens)
Output, base tier	$0.50	$0.50
Cache read, base tier	$0.01	$0.01
Cache write, base tier	$0.125 (5-minute)	$0.125
Higher tier, input / output	$0.50 / $2.50 (over 100K)	$0.20 / $0.75 (over 272K)
Higher tier, cache read / write	$0.05 / $0.625	2x cache rates
Batch	$0.05 / $0.25 up to 100K; $0.25 / $1.25 over	50% of Standard rates (Batch and Flex)

Sources: Anthropic’s pricing docs and OpenAI’s GPT-6 Luna model page.

OpenAI reprices “the full request” above 272K. Anthropic says “a prompt of over 100,000 tokens pays higher prices,” charged per request by prompt length.

Apply the pricing thresholds

1. Measure the input token count for each request.
2. If prompts stay below 100K tokens, the listed input and output rates are identical.
3. For prompts between 100K and 272K tokens, Haiku 5.5 costs 5x more for both input and output.
4. Above 272K tokens, both models use higher-tier pricing, but Luna remains cheaper.

Two examples, counted in each vendor’s own tokens:

• 150K-token prompt, 2K output
  • Haiku 5.5: 0.15 × $0.50 + 0.002 × $2.50 = $0.080
  • Luna: 0.15 × $0.10 + 0.002 × $0.50 = $0.016
• 300K-token prompt, 2K output
  • Haiku 5.5: 0.30 × $0.50 + 0.002 × $2.50 = $0.155
  • Luna: 0.30 × $0.20 + 0.002 × $0.75 = $0.0615

Anthropic says prompts up to 100K tokens made up around 90% of Haiku 4.5 requests. If your traffic has a similar distribution, most calls cost the same on either model. See Claude Haiku 5.5 pricing for more calculations.

Tokenizers move the threshold

Do not compare raw token counts without measuring both APIs. A token on one model is not necessarily a token on the other.

Haiku 5.5’s updated tokenizer produces about 30% more tokens than Haiku 4.5 for the same text, according to Anthropic. In the Hacker News launch thread, one commenter estimated that “modern Claude’s 100K tokens are about ~60-65K modern GPT tokens.” That is a community estimate, not a vendor figure.

For production routing:

1. Send representative prompts to both APIs.
2. Record each response’s usage object.
3. Compare input-token counts, output-token counts, latency, and cost.
4. Set routing thresholds from measured usage rather than character counts.

Specs

	Claude Haiku 5.5	GPT-6 Luna
API model ID	claude-haiku-5-5	gpt-6-luna
Context window	1M tokens	1,050,000 tokens
Max output	128K tokens (300K on Batch with a beta header)	128,000 tokens
Knowledge cutoff	June 2026	May 18, 2026
Effort levels	low, medium (default), high, xhigh, max	none, low, medium (default), high, xhigh, max
Thinking off	thinking: {"type": "disabled"} at high effort or below	none reasoning effort
Free chat access	Free Claude.ai users can select it	Free and Go users in the ChatGPT desktop app

Both models default to medium.

• On Haiku 5.5, configure effort with output_config.effort.
• On Luna, configure effort with reasoning.effort.
• Neither free chat route provides an API key, so neither can run a script, CI job, or agent.

See Claude Haiku 5.5 for free and GPT-6 Luna for free.

One caching difference matters for multi-step loops: on Haiku 5.5, changing top-level effort between requests invalidates the prompt cache. OpenAI says changing reasoning effort no longer breaks the cache on GPT-6 models. See GPT-6 prompt caching.

Benchmarks: Anthropic’s table and who ran each Luna cell

Every number below comes from Anthropic’s launch post and its system card. Anthropic compiled the table, but different parties ran different rows. Haiku 5.5 results are at max effort, mostly averaged over five trials.

Benchmark	Haiku 5.5	GPT-6 Luna	How the Luna cell was sourced
GDPval-AA v2.1 (Elo)	1620	1437	Run independently by Artificial Analysis
AA-Briefcase v1.1 (Elo)	1578	1336	Run independently by Artificial Analysis
OSWorld 2.1, offline subset	72.4%	48.9%	Run by Anthropic on the same 82 tasks via OpenAI’s API
Terminal-Bench 4.0	39.2%	16.4%	Public leaderboard, Codex CLI at max effort
FrontierCode 1.1 (Main)	46.4%	42.4%	Run by Cognition (Claude in Claude Code, GPT in Codex)
Chartography, no tools	46.4%	29.1%	As publicly reported by Surge AI

The rows share a benchmark name, not a single harness:

• Terminal-Bench uses Haiku 5.5 in Claude Code and Luna in the Codex CLI.
• OSWorld is the closest to a controlled comparison because Anthropic ran both models on the same tasks.
• OSWorld is still one vendor running a rival’s model.

No independent head-to-head exists yet. As of October 8, 2026, Artificial Analysis has no Haiku 5.5 page, and no Haiku 5.5 result was found on Vals, LMArena, SWE-bench, or Aider. AA’s leaderboard that day lists GPT-6 Luna (max) at an Intelligence Index of 38 and 129 output tokens per second.

For the full set of Haiku 5.5 results, see Claude Haiku 5.5 benchmarks.

Cost per attempt at each effort level

The launch table shows each model at max effort. Anthropic’s launch page also plots effort level against cost, including Luna on two charts.

OSWorld 2.1: offline subset

Partial-credit score / cost per attempt:

Effort	Haiku 5.5	GPT-6 Luna
Low	42.0% / $0.0695	19.2% / $0.038
Medium	53.3% / $0.1257	37.5% / $0.1261
High	61.3% / $0.1827	42.3% / $0.1435
Xhigh	67.6% / $0.2792	44.8% / $0.1672
Max	72.4% / $0.6111	48.9% / $0.205

GDPval-AA v2.1

Elo / cost per task:

Effort	Haiku 5.5	GPT-6 Luna
Low	1125 / $0.01167	1036 / $0.0038
Medium	1277 / $0.03042	1262 / $0.02
High	1420 / $0.08938	1344 / $0.03
Xhigh	1513 / $0.26725	1364 / $0.05
Max	1620 / $0.86593	1437 / $0.09

Luna costs less at every effort level on GDPval-AA and at low, high, xhigh, and max on OSWorld. At OSWorld medium, Haiku 5.5 is marginally cheaper: $0.1257 versus $0.1261.

Haiku 5.5 scores higher at every level. Compare models at matched spend instead of only comparing maximum scores:

• OSWorld: Haiku 5.5 at medium costs $0.1257 per attempt versus Luna’s $0.1261, while scoring 53.3% versus 37.5%. Haiku at medium also beats Luna at max (48.9%) for less money.
• GDPval-AA: Luna at max (1437, $0.09) and Haiku 5.5 at high (1420, $0.08938) cost about the same and are 17 Elo apart.

For computer-use workflows, Haiku 5.5 buys more score per dollar in Anthropic’s run. For knowledge work, the models are close at similar spend, while Luna’s lower floor is useful when good-enough output is the goal.

Routing: which model for which workload

Anthropic positions Haiku 5.5 for “narrowly scoped tasks” and says Sonnet 5.5 and Opus 5.5 “remain better choices for complex agentic coding tasks.” Both models here are built for volume.

Workload	Start with	Why
Classification, routing, extraction under 100K tokens	Either; test both	Identical price; choose based on accuracy for your labels
Long-document Q&A, 100K to 272K tokens	GPT-6 Luna	Base rate holds to 272K; Haiku 5.5 is 5x above 100K
Browser and desktop automation	Claude Haiku 5.5	Higher OSWorld score at matched cost
Subagents under Opus 5.5 or Sonnet 5.5	Claude Haiku 5.5	Claude Code model: haiku frontmatter uses Haiku 5.5 on the Anthropic API
Lowest cost per call	GPT-6 Luna	Cheapest floor: $0.038 on OSWorld and $0.0038 on GDPval-AA at low
Knowledge-work drafting at high effort	Either; test both	Within 17 Elo at matched spend on GDPval-AA

See Claude Haiku 5.5 in Claude Code for subagents and GPT-6 Luna for high-volume API workloads for high-QPS pipelines.

Coming from Haiku 4.5? Haiku 5.5 vs Haiku 4.5 lists the five new 400 errors.

Test both side by side in Apidog

The comparison that matters runs on your prompts. In Apidog, create one project with two requests that use the same prompt.

1. Store ANTHROPIC_API_KEY and your OpenAI key as environment variables.
2. Add the Haiku 5.5 request below, then add the Luna request using the same prompt and effort.
3. Assert that Haiku’s stop_reason is not "refusal" because there is no server-side fallback.
4. Add a token-budget assertion based on the response usage.
5. Run both models at low, medium, and high effort.
6. Save responses and compare token usage, latency, cost, and output quality.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-5-5",
    "max_tokens": 2048,
    "thinking": {"type": "adaptive"},
    "output_config": {"effort": "medium"},
    "messages": [
      {
        "role": "user",
        "content": "Classify this support ticket as billing, bug, or feature request: The export button returns a 500 error since this morning."
      }
    ]
  }'

Leave out temperature, top_p, and top_k: non-default values return a 400 on Haiku 5.5.

For a walkthrough, see how to use the Claude Haiku 5.5 API. For eval sets, see testing LLM applications.

FAQ

Are Claude Haiku 5.5 and GPT-6 Luna the same price?

Yes, for Haiku prompts up to 100K tokens: $0.10/$0.50, $0.01 cache reads, and $0.125 cache writes. Above that, Haiku 5.5 costs $0.50/$2.50, while Luna stays at base pricing until 272K.

Which model scores higher?

Haiku 5.5 scores higher on every shared row of Anthropic’s launch table. Anthropic compiled that table, and no independent head-to-head exists yet.

Which is cheaper per task?

Usually Luna: it costs less at every effort level on Anthropic’s GDPval-AA chart and at four of five effort levels on OSWorld. At OSWorld medium, Haiku 5.5 is $0.0004 cheaper. At matched spend, Haiku 5.5 leads on OSWorld, while GDPval-AA is close.

Is either one free?

Only in chat interfaces. Free Claude.ai users can select Haiku 5.5, and Free and Go users get Luna in the ChatGPT desktop app. Neither provides an API key. See Claude Haiku 5.5 for free.

Pick by prompt length, then test

Start with prompt length:

• Under 100K tokens: price is a tie. Test accuracy on your real tasks.
• From 100K to 272K tokens: Luna’s threshold saves money on every call.
• Over 272K tokens: Luna remains cheaper at the higher tier.

Then test both models on production-like requests at two or three effort levels. Download Apidog, put both requests in one project, and use the returned usage values to make the routing decision.