← The Polylog AI Intelligence Brief
Morning Edition · Thursday, October 8, 2026Published at 2:25 AM EDT · New York
Anthropic prices Claude Haiku 5.5 at $0.10 per million input tokens, exactly matching OpenAI's GPT-6 Luna
With Sonnet 5.5 and GPT-6 Sol both listed at $2 and $10 per million tokens, Anthropic and OpenAI now charge identical prices at two tiers, and both small models list below DeepSeek V4.1 Flash's off-peak rate.
Anthropic released Claude Haiku 5.5 on October 7 and describes it as its fastest, cheapest and most capable small model. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens. Above that length the price rises to $0.50 and $2.50. The model has a 1 million-token context window and supports outputs of up to 128,000 tokens. It is also the first Haiku-class model with an adjustable effort setting. A Russian-language channel that covers releases describes the target workloads as high-volume classification, automation, simple agents and API traffic.
Anthropic's reported capability gains are large, and the comparisons need context. Haiku 5.5 scored 72.4% on the OSWorld 2.1 offline subset, compared with 15.7% for Haiku 4.5. On Terminal-Bench 4.0 it scored 39.2%, while Haiku 4.5 scored 0.0%. A zero for the predecessor suggests that Terminal-Bench 4.0 was built to be hard enough to separate current models. The jump therefore says as much about how difficult the benchmark is as about Haiku 5.5. All of these figures come from the vendor, and no independent reproduction has been published yet.
On cost, AI Post reports that Haiku 5.5 is about 75% cheaper to run than Haiku 4.5. Haiku 4.5 listed at $1 and $5 per million tokens. Short-prompt list prices are therefore 90% lower and long-prompt prices 50% lower, so a blended figure near 75% is plausible for mixed traffic. The Chinese tech outlet 36Kr says the release pushes small-model competition into a "1-cent era".
The prices also match OpenAI's. OpenAI's GPT-6 Luna lists at $0.10 and $0.50, and GPT-6 Sol lists at $2 and $10. That is the same as Anthropic's Sonnet 5.5 at $2 and $10. Opus 5.5, which Anthropic says performs at the level of Claude Fable 5.1 on most work, costs $4 and $20. For comparison, DeepSeek's V4.1 Flash lists off-peak at $0.15 for uncached input and $0.60 for output.
What this means
The two leading US labs now price their small and mid tiers at identical rates. That points to competition that matches prices tier by tier rather than one lab undercutting the other. The pattern resembles the 2014 cloud price cuts, when Amazon Web Services, Microsoft Azure and Google Cloud matched each other's reductions and then settled into stable oligopoly pricing. The gainers are application companies and agent builders that run many subagent calls, because their input costs drop by roughly an order of magnitude. Lab revenue per token falls, so revenue growth must now come from volume. The new fact here is that US small models now list below DeepSeek's off-peak price per token. That weakens the cost argument for the Chinese open-weight stack, though not the argument for weights that can be self-hosted. The desk had marked the price-war trend as weakening. This evidence argues against that call. The desk's read is that Google will price its next Flash-class model at or below $0.10 and $0.50 within weeks, and that DeepSeek will respond with another off-peak cut. That depends on Google not having already decided to compete on bundled distribution instead of list price. If Google leaves Flash pricing unchanged through the quarter, this read is wrong.
Part of a tracked trend
Frontier Model Price War
Frontier API vendors increasingly compete on strategic price-cutting rather than pure capability, repeatedly launching or repricing models below prevailing rates to grab share and compressing industry inference margins; expect recurring below-rival pricing moves.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI's 722 machine-written math manuscripts land on mathematicians to verify as Terence Tao warns of a rush to "flatten" open problems
- GPT-6 reaches all ChatGPT users with interactive "Intelligent UI" as OpenAI adds 512 MB audio uploads for paid plans
- Nvidia and Microsoft open pre-orders for Arm-based RTX Spark Windows laptops with up to 128 GB of unified memory
- White House gathers more than $1 billion in private pledges from AMD, OpenAI and Anthropic for its Genesis Mission science program
- Google and Cambridge's VeriHarness uses the generating model to audit its own agent rollouts and adds about 6 points on long-horizon tasks
- Google releases EmbeddingGemma 2, a 740-million-parameter open embedder for text, code, images and audio under Apache 2.0
- MIT's open-weight FleXray segments 60 anatomical structures in X-rays after training only on synthetic images rendered from CT scans
Comments
0No comments yet.