GPT-6.1 Sol, Argon: $2/$10 per 1M Tokens — AI Digest

GPT-6.1 Sol, Argon: $2/$10 per 1M Tokens — AI Digest

OpenAI prices GPT-6.1 Sol at $2/$10 per 1M tokens, Google opens Gemini 4 Argon to a narrow group, and Claude Sonnet 5.5 holds $2/$12 while ranking #3 in Agent Arena.

Top stories

GPT-6.1 Sol: $2/$10 per 1M tokens, $0.72 per task against Astra's $3.26

OpenAI's GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, one fifth of GPT-6 Astra's $10/$50. At its DevDay on September 29, 2026, OpenAI pitched the model as near-Astra intelligence for a fifth of the price. Artificial Analysis then measured $0.72 per Intelligence Index task at maximum effort, against $1.04 for GPT-6 Sol and $3.26 for Astra. Its explanation: fewer turns and cheaper cache reads, not fewer generated tokens. OpenAI also fixed an image-encoding bug, which added one Index point to GPT-6 Luna. Sam Altman called Sol the company's fastest-growing model.

OpenAI Dots: always-on agents with their own cloud computer

OpenAI's Dots are persistent personal agents that message you first and connect to more than 4,000 apps. Each dot runs on GPT-6 Astra with its own cloud computer, and users set what it may do alone, what needs approval and what is off limits. Ben's Bites describes a Dot as a never-ending orchestrator chat that spins up Codex and ChatGPT Work tasks and reports back; Pro plans get one. Ethan Mollick, writing in One Useful Thing, groups Dots with Meta's Muse as "Clawlikes", successors to OpenClaw, and says his own agents increasingly catch his mistakes rather than the reverse. He adds that Muse currently tops the App Store.

Gemini 4 Argon: Google claims 13 of 19 benchmark wins, access stays restricted

Google DeepMind's Gemini 4 Argon lists at $4/$20 per 1M tokens, halved to $2/$10 by an introductory discount with no announced end date. Access starts with government users and vetted cyber defenders in the Fairwind Program, according to Google DeepMind and The Rundown AI (October 1, 2026). Google says Argon leads 13 of 19 published benchmarks against GPT-6 Astra and Claude Opus 5.5 and scores 77.9% on DeepSWE versus 74.2% for Opus 5.5, with output of up to 1M tokens. Artificial Analysis gives it 53 on its Intelligence Index, level with Astra.

Claude Sonnet 5.5: unchanged $2/$12 pricing, 30% faster, #3 in Agent Arena

Anthropic released Claude Sonnet 5.5 on September 28, 2026 at Sonnet 5's price of $2/$12 per 1M tokens, claiming over 30% more speed and up to 30% lower cost per task. Anthropic calls it a clear upgrade over Sonnet 5. Vals AI lists a 1M-token context window and up to 128K output. In Agent Arena the model debuted at #3 and first in the Chat category, at $2.74 per task against $1.58 for Opus 5.5. One caveat: at maximum effort Artificial Analysis recorded about 193K output tokens per task, the most it has measured. Anthropic staff advise against running Sonnet at max effort.

Decision models: Cloudflare, Perplexity and OpenAI turn classification into its own layer

Three vendors shipped decision models in one week: fast models that pick one answer from a fixed list. A decision model is a small model that takes a question plus fixed answer options and returns probabilities over them instead of free text. OpenAI launched the Decisions API on GPT-6 Luna for near-instant classification and routing. Perplexity released pplx-decider-v1-27b, built on Qwen3.8-27B, at $0.04 per million input tokens, free output and open weights. Cloudflare's two Clef models come with Apache 2.0 weights. Skeptics such as @mervenoyann call the category a rebrand of zero-shot classifiers.

Numbers and facts

Our price table for this week's models (per 1M input / output tokens, from the sources cited in this issue):

ModelInput / outputNote
GPT-6.1 Sol$2 / $10$0.72 per task (Artificial Analysis)
Claude Sonnet 5.5$2 / $121M context, 128K output
Gemini 4 Argon$4 / $20$2 / $10 with the 50% promo; $1.99 per task
GPT-6 Astra$10 / $50$3.26 per task
GPT-6 Astra Ultrafast$60 / $300six times the normal price, up to 300 tokens/s
Solar Mini 4$0.10 / $0.4035B total / 3B active, closed weights

Different views: is Gemini 4 Argon the new frontier leader?

Argon tops the Vals Index at 68.9% and places in the top five on 20 of 22 Vals benchmarks, yet Epoch AI ranks Opus 5.5 first on its ECI at 167, with Sonnet 5.5 roughly level with Fable 5.1 at 165. Those are different test suites, not interchangeable leaderboards.

For. Vals puts Argon first on its aggregate index, and Google says new Gemini revisions now go through weeks of testing by thousands of internal engineers. Argon also took first place in Text Arena.

Neutral. Artificial Analysis scores Argon level with Astra at 53, but the savings come from price: Argon spends 62K output tokens per task against Astra's 27K.

Against. A circulated Bloomberg report blamed weak coding on anonymous insiders; a senior DeepMind engineer reportedly rejected that account. On Harvey's legal benchmark Argon scored 19.6% against 25.42% for Muse Spark 1.2, and other commentators suspect tuning to preference data. Real-world coding quality remains disputed.

Tools and techniques

In brief