September 2026 was a crowded month. OpenAI released GPT-6 Astra and GPT-6.1 Sol. Anthropic released Claude Fable 5.1, Opus 5.5 and Sonnet 5.5. Google released Gemini 3.8 Flash and announced Gemini 4 Argon. Each company says its model leads.
Here’s how they compare on the things you can check, price and access, and a way to choose that doesn’t depend on whose launch post you read last.
The models side by side
List prices per million tokens, as published on 6 October 2026:
| Company | Model | Input / output price | Where you can use it |
|---|---|---|---|
| OpenAI | GPT-6 Astra | $10 / $50 | ChatGPT Plus, Pro, Business and Enterprise; OpenAI API; Microsoft Azure; AWS Bedrock |
| OpenAI | GPT-6.1 Sol | $2 / $10 | ChatGPT Work and Codex on paid plans; OpenAI API |
| Anthropic | Claude Fable 5.1 | $10 / $50 | Claude apps (paid plans), Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Anthropic | Claude Opus 5.5 | $4 / $20 | Claude apps (paid plans), Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Anthropic | Claude Sonnet 5.5 | $2 / $10 | Claude apps, Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Gemini 4 Argon | $2 / $10 at launch, then $4 / $20 | Cyber defenders in Google’s Fairwind Program, for now | |
| Gemini 3.8 Flash | $0.75 / $3.75 until 31 Dec 2026 | Gemini API, Google AI Studio, Gemini Enterprise, Gemini app (Google AI Pro and Ultra) |
Caching and batch discounts cut these further, and prices change often, so check the vendor’s page before you set a budget.
Three things stand out:
- The $2 / $10 tier is crowded. GPT-6.1 Sol and Claude Sonnet 5.5 sit there today, and Gemini 4 Argon will start there. That’s where most business apps should begin testing.
- Google is the cheapest place to start. Gemini 3.8 Flash costs well under half of that, at least until its introductory price ends on 31 December 2026.
- The top tier costs five times more. GPT-6 Astra and Claude Fable 5.1 are both $10 / $50. OpenAI itself says GPT-6.1 Sol comes close to GPT-6 Astra on coding, computer use and professional work at one-fifth of Astra’s standard API prices.
What about benchmarks?
Every company publishes results on tests it chooses, and the tests rarely overlap. Where they do:
- On DeepSWE v1.1 (real-world software engineering), OpenAI reports 74.1% for GPT-6 Astra and Google reports 77.9% for Gemini 4 Argon.
- On AutomationBench, OpenAI reports 41.4% for GPT-6 Astra and Google reports 51.3% for Gemini 4 Argon.
- TechCrunch reports Google’s claim that Argon scored significantly higher than GPT-6 Astra and Anthropic’s Fable and Opus models across a range of benchmarks. Anthropic publishes its own evaluations with each model.
Each company ran these tests itself, under its own setup. Use benchmarks to build a shortlist, not to make the final call.
How to choose for your app
- Write down the job. “Answer customer questions from our help pages.” “Pull 12 fields from supplier invoices.” “Review every pull request.” Different jobs suit different models.
- Shortlist one model from each company in the right tier. For most jobs that means Gemini 3.8 Flash, Claude Sonnet 5.5 and GPT-6.1 Sol.
- Test on your own data. Use 30 to 50 real examples and score them the way your team would: right or wrong, complete or not, safe or not. Track cost per task and response time too.
- Check where it can run. If your company already uses AWS, Google Cloud or Microsoft, you may be able to use a model inside that account. Claude is offered on all three; GPT-6 Astra on Azure and AWS Bedrock; Gemini on Google Cloud.
- Plan a fallback. In June 2026, Anthropic had to suspend its newest models for nearly three weeks after US export controls were applied to them. Models also get retired: Google set 1 June 2026 as the shutdown date for Gemini 2.0 Flash. Design your software so switching models is a settings change.
- Re-test every few months. New models arrive almost monthly. Re-running your test set tells you in an afternoon whether switching is worth it.
Where each company stands out right now
Based on what each has shipped and published:
- Google: the lowest entry price (Gemini 3.8 Flash), real-time voice models (Gemini 3.8 Live), image generation (Nano Banana), and Gemini built into Google’s own tools, including Sheets.
- OpenAI: computer use. OpenAI reports 72.6% for GPT-6 Astra on OSWorld 2.0, which matters for agents that operate websites and desktop software.
- Anthropic: a clear ladder from Haiku 4.5 to Fable 5.1, a 1M-token context window on Sonnet, Opus and Fable, and availability on all three big clouds.
The bottom line
There’s no single best model, and anyone who says otherwise is usually selling one. Pick by task, test on your own data and keep a fallback. Most apps end up using two models: a fast, cheap one for simple steps and a stronger one for the hard parts.
For the detail on each family, read our guides to Gemini models, Gemini 4 Argon and Claude models.
FAQ
Which AI model is cheapest for business use?
Of the models in this comparison, Gemini 3.8 Flash, at $0.75 per million input tokens and $3.75 per million output tokens until 31 December 2026.
Which is best for coding?
The vendors’ own numbers disagree. On DeepSWE v1.1, Google reports 77.9% for Gemini 4 Argon and OpenAI reports 74.1% for GPT-6 Astra, and Anthropic positions Claude Opus 5.5 for long-running agentic coding. Test on your own codebase.
Can we use more than one model in the same app?
Yes, and we usually recommend it. Send simple steps to a cheap model, hard ones to a stronger model, and keep a fallback from another company.
Should we fine-tune a model?
Usually not as a first step. Start with clear instructions and answers drawn from your own documents (retrieval), and fine-tune only if tests show a gap that those can’t close.
Not sure which fits?
We test models on a sample of your real data and recommend one, with the cost per task. Get your AI readiness report or see our AI services.
Sources
- OpenAI: GPT-6 Astra and Introducing GPT-6.1 Sol
- Claude Platform Docs: Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1
- Anthropic: Redeploying Claude Fable 5
- Google: Gemini 4 Argon, Gemini 3.8 Flash, Gemini API models
- TechCrunch: Google releases Gemini 4 Argon