{"slug": "ai-labs-raise-prices-without-raising-prices", "title": "AI labs raise prices without raising prices", "summary": "OpenAI's GPT Image 2.6 and Anthropic's Claude Fable models charge the same per-token prices as their predecessors but consume far more output tokens for identical tasks, according to measurements published by Playcode, which found GPT Image 2.6 used 1,449 output tokens versus 439 for GPT Image 2.5 on the same 1024x1024 medium prompt — a 3.3x cost increase from $0.013 to $0.044. Playcode also measured Claude Fable 5.1 spending 2.3x the output tokens of Fable 5 on its MacBook SVG benchmark, and reported that Anthropic's tokenizer change made the same text count about 30% more tokens in Opus 4.8 versus Opus 4.6. In a November 9 update, Playcode reported Claude Fable 5.2 at $8/$40 per million tokens versus $10/$50 for Fable 5.1, yet the same drawing cost $3.05 versus $0.90 — 3.4x more — because Fable 5.2 spent 76,158 output tokens against 17,933.", "body_md": "TL;DR\n\nSame price per token, more output tokens. GPT Image 2.6 spends 3.3x the output tokens of 2.5 on the same picture. Anthropic changed its tokenizer, and the same text now comes out about 30% more tokens. Update, November 9: Claude Fable 5.2 is a fifth cheaper per token than Fable 5.1, and the same drawing cost 3.4x as much.\n\nNew model, same price, everything looks great, you'd say, let's upgrade. But nope! There's a new trend now, cheating with output tokens. The price looks the same, but they've found a way to increase the price by increasing output tokens. Anthropic does the same thing too. Same price, more output tokens. In the end, they earn more. And you won't notice right away, not until you start comparing prices on the same task.\n\nHere's how I noticed. OpenAI released GPT Image 2.6 yesterday and I told my team to switch everything to it. They came back and said Sunburst 2.6 costs three times more. I said, are you sure the price went up threefold? Let me take a look too. Sunburst 2.5: input 8, output 30, cached 2. Same for Sunburst 2.6.\n\nI thought they were mixing something up. Or does it use more tokens?\n\nIt does. This is the usage OpenAI returned for the same prompt at 1024x1024, medium:\n\n```\ngpt-image-2.5-sunburst   \"output_tokens\": 439    # $0.013\ngpt-image-2.6-sunburst   \"output_tokens\": 1449   # $0.044\n```\n\nWhoa, that's a hell of a difference. We ran two more prompts to be sure and got the same numbers.\n\n| Request | GPT Image 2.5 | GPT Image 2.6 | 2.6 costs | \n|---|---|---|---|\n| 1024x1024, medium | 439 tokens | 1,449 tokens | **3.3x** | \n| 1536x1024, medium | 343 tokens | 1,132 tokens | **3.3x** | \n| 2.6 at its lowest quality vs 2.5 at medium | 439 tokens | 824 tokens | **1.9x** | \n\nThe 2.6 picture is a bit sharper, more contrast. Three times better? I don't think it's worth it.\n\n So what's the method they use? So far I see two. The first one is inflating output tokens. GPT Image 2.6 is one example, and Claude Fable 5.1 is another: on the same task it spent 2.3x the output tokens of Fable 5, same price per token (we measured it in our [MacBook SVG benchmark](https://playcode.io/blog/macbook-svg-benchmark)). \n\n The second one, Anthropic already did this once, is changing the tokenizer so it counts more. Opus 4.8 costs the same as Opus 4.6, but the same text comes out about 30% more tokens (full count in our [tokenizer post](https://playcode.io/blog/real-price-of-frontier-models)). \n\nIf you don't know how to raise the price, but you need to, AI labs have figured it out.\n\nWhat we did: 2.5 stays the default in Playcode. We use 2.6 only where it's worth it, for the sharper look or for GIFs and SVGs, which 2.5 can't do.\n\n Check it yourself, it costs about six cents. Run the same prompt on both models and look at `usage.output_tokens`. We got 439 and 1449. \n\n```\nfor m in gpt-image-2.5-sunburst gpt-image-2.6-sunburst; do\n  curl -s https://api.openai.com/v1/images/generations \\\n    -H \"Authorization: Bearer $OPENAI_API_KEY\" \\\n    -H \"Content-Type: application/json\" \\\n    -d \"{\\\"model\\\":\\\"$m\\\",\\\"size\\\":\\\"1024x1024\\\",\\\"quality\\\":\\\"medium\\\",\\\"prompt\\\":\\\"A flat vector illustration of a red fox reading a book under a tree at night\\\"}\" \\\n  | jq -r --arg m \"$m\" '\"\\($m): \\(.usage.output_tokens) output tokens\"'\ndone\n```\n\nSame price, more output tokens. In the end, they earn more.\n\n## Update, November 9: Claude Fable 5.2\n\nAnthropic released Claude Fable 5.2 today, at $8 / $40 per million tokens. Fable 5.1 was $10 / $50.\n\nI definitely saw a jump in price from Fable 5 to Fable 5.1, and now surely 5.2 will be more expensive too, I'm predicting that.\n\n Same MacBook drawing from our [benchmark](https://playcode.io/blog/macbook-svg-benchmark), same effort, high: \n\n| Model | Price per M tokens | Output tokens | Cost | Time | \n|---|---|---|---|---|\n| Fable 5 | $10 / $50 | 5,383 | $0.27 | 1.1 min | \n| Fable 5.1 | $10 / $50 | 17,933 | $0.90 | 3.8 min | \n| Fable 5.2 | $8 / $40 | 76,158 | **$3.05** | 12.5 min | \n\nIt is: 3.4x. At xhigh and max, 5.2 spent all 128,000 output tokens on thinking and returned nothing, $5.12 each. 5.1 finished both. Given 300,000 on the Batch API, 5.2 finished max with 250,082 tokens: $10.00 at the standard price, where 5.1 needed 99,072. On the pelican, 5.2 wrote 2.6x to 3x the tokens of 5.1 at every level.\n\nOpus 5.5 did it in September: $4 / $20 against Opus 5's $5 / $25, and 1.8x the output tokens at xhigh on the same drawing, $1.72 against $1.18. At max it can't finish inside 128,000 at all. When we made it our default, I said it'll burn through money fast.\n\n**Sources:** OpenAI's price list ([developers.openai.com/api/docs/pricing](https://developers.openai.com/api/docs/pricing)) and image generation guide ([developers.openai.com/api/docs/guides/image-generation](https://developers.openai.com/api/docs/guides/image-generation)), read on November 5, 2026. Token counts are the `usage` fields of our own Images API calls on November 5, 2026. The Anthropic and Fable figures come from our earlier measurements linked above. Claude Fable 5.2's price: Anthropic's pricing page ([platform.claude.com/docs/en/about-claude/pricing](https://platform.claude.com/docs/en/about-claude/pricing)), read on November 9, 2026; its tokens, costs and times come from the [MacBook SVG benchmark](https://playcode.io/blog/macbook-svg-benchmark) runs of the same day.", "url": "https://wpnews.pro/news/ai-labs-raise-prices-without-raising-prices", "canonical_source": "https://playcode.io/blog/ai-labs-raise-prices-without-raising-prices", "published_at": "2026-11-09 22:15:58+00:00", "updated_at": "2026-11-09 22:51:43.642729+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "generative-ai", "ai-products"], "entities": ["OpenAI", "GPT Image 2.6", "GPT Image 2.5", "Anthropic", "Claude Fable 5.2", "Claude Fable 5.1", "Claude Opus 4.8", "Playcode"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/ai-labs-raise-prices-without-raising-prices", "markdown": "https://wpnews.pro/news/ai-labs-raise-prices-without-raising-prices.md", "text": "https://wpnews.pro/news/ai-labs-raise-prices-without-raising-prices.txt", "jsonld": "https://wpnews.pro/news/ai-labs-raise-prices-without-raising-prices.jsonld"}}