Newest GPT vs Claude Models: Price and Performance 2026
The short answer
On capability, Claude is ahead in every tier; on cost per task, GPT is usually cheaper because it uses fewer tokens. List prices now match in the mid and small tiers, and Claude’s flagship costs less than half of OpenAI’s. Prices are per million tokens (input / output) from each vendor’s pricing page; scores and cost per task are from Artificial Analysis (AA Intelligence Index, max effort), read October 14, 2026:
| Tier | Model | Price | AA Index | AA cost per index task | Output speed |
|---|---|---|---|---|---|
| Flagship | Claude Opus 5.5 | $4 / $20 | 58 | $5.98 | ~95 tok/s |
| Flagship | GPT-6 Astra | $10 / $50 | 53 | $3.26 | ~60 tok/s |
| $2 tier | Claude Sonnet 5.5 | $2 / $10 | 56 | $5.46 | ~136 tok/s |
| $2 tier | GPT-6.1 Sol | $2 / $10 | 52 | $0.72 | ~59 tok/s |
| Small | Claude Haiku 5.5 | $0.10 / $0.50 (≤100K prompt) | 43 | $0.21 | ~237 tok/s |
| Small | GPT-6 Luna | $0.10 / $0.50 (≤272K prompt) | 38 | $0.07 | ~135 tok/s |
Claude Fable 5.1 ($10/$50, AA 53) sits above Opus as Anthropic’s long-horizon model; for most work Opus 5.5 now scores higher at 40% of the price. Full table: current API prices.
Tier by tier
Flagship: Claude Opus 5.5 is the better buy. It is cheaper per token than GPT-6 Astra and scores 5 points higher. Astra’s cost per task is lower at max effort ($3.26 vs $5.98) because Opus at max emits very long reasoning; at medium effort Opus scores 51 for $1.34 per task. Pay for Astra when you are inside the OpenAI stack (Codex, ChatGPT Work) or need its speed tiers.
$2 tier: Sonnet 5.5 for quality, GPT-6.1 Sol for volume. Same list price, different shape. Sonnet 5.5 scores higher and is faster. GPT-6.1 Sol reaches 52 while spending far fewer tokens, so it costs about one seventh as much per AA task; at xhigh effort it scores 51 for $0.39. Both have $0.10 cached input.
Small: Haiku 5.5 for speed and quality, Luna for the lowest bill. Haiku 5.5 is smarter and nearly twice as fast; Luna is about three times cheaper per task. Watch long prompts: Haiku 5.5 reprices the whole request to $0.50/$2.50 above 100K tokens, Luna only to $0.20/$0.75 above 272K. Detail in Claude Haiku 5.5 vs GPT-6 Luna vs Gemini 3.8 Flash.
Cost levers that change the answer
- Effort setting. The same model can cost 5–10x more per task at max effort than at medium. Benchmark at the effort you will run.
- Prompt caching. Cached input is $0.10 on GPT-6.1 Sol and Claude Sonnet 5.5, $0.20 on Opus 5.5 and $1.00 on GPT-6 Astra.
- Batch. Both vendors take 50% off batch jobs.
- Speed tiers. Anthropic’s fast mode for Opus 5.5 is $8/$40; OpenAI’s Fast mode is 2x and Ultrafast 6x the standard price.
Deeper head-to-heads: Claude Sonnet 5.5 vs GPT-6.1 Sol and GPT-6.1 Sol vs GPT-6 Astra vs Claude Opus 5.5. Google’s new flagship is covered in Gemini 4 Argon now available.
Last verified: October 14, 2026.