# Anthropic and OpenAI price their near-flagship models at $2 and $10 per million tokens while open-weight Qwen runs on a gaming GPU

Claude Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, ahead of Opus 5.5's 66.4%, at an unchanged list price. Most of the cost reduction comes from fewer tokens and cheaper cache reads.

- Published: 2026-10-05T06:00:43.302Z
- Canonical: https://polylog.news/ai/2026-10-05/anthropic-and-openai-price-their-near-flagship-models-at-2-a
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Anthropic News (Claude Sonnet 5.5)](https://www.anthropic.com/claude-sonnet-5-5), [Anthropic News (Claude Opus 5.5)](https://www.anthropic.com/claude-opus-5-5), [Hacker News (Strata)](https://github.com/Niko1221/Strata), [The Next Web](https://thenextweb.com/news/openai-gpt-6-1-sol-price-astra-devday), [The Decoder](https://the-decoder.com/anthropics-claude-sonnet-5-5-nearly-matches-opus-5-5-on-benchmarks-while-costing-up-to-30-percent-less-per-task/)

Within one week, Anthropic and OpenAI each released a second-tier model that their own evaluations place within a few points of their flagship. Both set the same list price. [Anthropic released Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) on September 28 at [$2 per million input tokens and $10 per million output tokens](https://www.marktechpost.com/2026/09/28/anthropic-releases-claude-sonnet-5-5-70-6-on-terminal-bench-4-0-at-the-same-2-10-price/), the same rates as Sonnet 5. The next day, at its DevDay developer event, [OpenAI shipped GPT-6.1 Sol](https://thenextweb.com/news/openai-gpt-6-1-sol-price-astra-devday) at $2 input, $0.10 cached input and $10 output. Those rates are one-fifth of what OpenAI charges for its flagship, GPT-6 Astra.

The gains Anthropic reports are large. According to [The Decoder's summary of Anthropic's figures](https://the-decoder.com/anthropics-claude-sonnet-5-5-nearly-matches-opus-5-5-on-benchmarks-while-costing-up-to-30-percent-less-per-task/), Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5 and above Opus 5.5's 66.4%. On CursorBench 4.0 it scores 55.5%, against 34.1% for Sonnet 5 and 57.8% for Opus 5.5. It scores 1844 on GDPval-AA against Opus 5.5's 1846, and 80.1% on OSWorld 2.1 against 81.8%. These are vendor numbers. The only independent result in hand is the [Vals Index](https://www.vals.ai/models/anthropic_claude-sonnet-5-5), which ranks Sonnet 5.5 second of 43 models at 67.04%.

Opus 5.5 arrived on September 22. [Anthropic says](https://www.anthropic.com/claude-opus-5-5) it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. List prices fell 20%, to [$4 input and $20 output per million tokens](https://the-decoder.com/claude-opus-5-5-matches-fable-5-1-at-40-percent-lower-cost-as-anthropic-promises-to-fix-claudish-writing/), and cache reads fell 60%, to $0.20. [One breakdown](https://claudefa.st/blog/models/claude-opus-5-5) estimates that most of the 40% saving comes from cheaper cache reads in long agent sessions. By that estimate, a short chat with little cached context saves closer to 20%.

OpenAI reports on the same basis, cost per completed task. At maximum effort, [GPT-6.1 Sol scores 57.0% on Terminal-Bench at $5.47 per task](https://thenextweb.com/news/openai-gpt-6-1-sol-price-astra-devday), against 27.6% at $12.18 for GPT-6 Sol. The two vendors report different Terminal-Bench configurations, so the Sol and Sonnet scores cannot be compared directly.

Open-weight models are also getting cheaper to run, which adds competitive pressure on these API prices. Alibaba's Qwen team released [Qwen3.8-Flash-Next](https://huggingface.co/Qwen/Qwen3.8-Flash-Next) on August 26 as a preview of its Qwen4 architecture. It is a sparse MoE with 125B main parameters, plus 51B parameters of N-gram embeddings, and it activates about 6B parameters per token. It supports a 262K-token context. The Qwen team says training [cost about one-ninth as much as Qwen3.7-Plus](https://qwen.ai/blog?id=qwen3.8-flash-next). An open-source engine called [Strata](https://github.com/Niko1221/Strata) keeps the most-used experts in GPU memory, holds the full expert set in system RAM and serves the N-gram table from SSD. Its developer reports [about 100 tokens per second on a single RTX 4090](https://www.promptzone.com/arlo_girard/strata-runs-qwen-125b-on-rtx-4090-at-100ts-550d), and other users report 40 to 60 tokens per second on an RTX 4080 Super. [Unsloth's documentation](https://unsloth.ai/docs/models/qwen3.8-next) says the model outperforms Claude 4.6 Opus. That comparison is with a model two Anthropic generations old, and no independent lab has reproduced it.

## What this means

The pricing competition has moved away from list rates, which did not change for Sonnet, and toward tokens, tool calls and cache reads per completed task. That favors vendors that control both the model and the serving stack. The exposed parties are the flagship tiers. OpenAI now sells a model it describes as near-Astra for one-fifth of Astra's rates, and Anthropic sells a Sonnet that scores 1844 against Opus 5.5's 1846 on GDPval-AA. Labs and enterprises expected price cuts and have already built them into their plans. The new fact is that the second tier now performs within a few points of the flagship on agentic evals, so the premium for the top tier needs a new justification. The consensus reads this as margin compression. The 2024 precedent points the other way: when GPT-4o mini cut per-token costs, usage volume rose by more than prices fell, which supported demand for compute. The likelier outcome is that within months both labs reposition or reprice Astra and Fable around harder long-horizon work, and that total inference spend keeps rising. That read depends on enterprise agent workloads continuing to grow. It would be proved wrong if enterprises continue paying flagship rates for routine work, or if a vendor raises rates on its second tier.

## What to watch

- Whether independent harnesses such as Vals and Artificial Analysis reproduce Sonnet 5.5's Terminal-Bench 4.0 gap over Opus 5.5. A second-tier model beating its own flagship is unusual. Independent confirmation would show the gap is real and not an artifact of Anthropic's harness.
- Any repricing or repositioning of GPT-6 Astra or Claude Fable. A cut would show that the labs cannot sustain a fivefold price premium on their flagship models.
- Independent benchmarks of Qwen3.8-Flash-Next under Strata's split across GPU, RAM and SSD. Offloading experts usually costs accuracy or speed on long contexts. That cost decides whether local inference can compete with $2/$10 APIs for agent work.