Gemini 4 Argon is GA at $2/M input, but the price is introductory
Google's Gemini 4 Argon went generally available on October 13 at $2 input and $10 output per million tokens. The rate is introductory, so agent and trading-research teams should model the post-promo price before committing.
By TRAGenX Desk
Google's first Gemini 4 model, Argon, reached general availability on October 13, 2026. According to Cryptopolitan's report, it costs $2 per million input tokens and $10 per million output tokens, with cached input about 95% cheaper at $0.10 per million.
What the pricing actually says
The headline number is lower than the frontier alternatives named in the same report: Claude Opus 5.5 at $4 input / $20 output, and OpenAI's GPT-6 Astra at $10 input / $50 output. Mashable's per-task estimate puts Argon at roughly 60% of GPT-6 Astra's cost on the Artificial Analysis Intelligence Index.
The catch is in the calendar. The rates are introductory and run until November 30, after which API pricing doubles. At $4/$20 Argon would match Opus 5.5's list price, so the discount currently on offer is a launch promotion, not a durable price gap.
Capability: cheaper is not the same as better
Google says Argon tops the Vals Index, which covers finance, coding, legal and tax tasks. In its own comparison table it leads in 14 of 18 results. On the independent Artificial Analysis Intelligence Index, Argon scores 53 on its High setting. That ties GPT-6 Astra but trails Claude Opus 5.5 and Sonnet 5.5 by about five points.
Vendor tables and third-party indexes answer different questions. Neither tells you how a model behaves on your tool schemas, your retry logic or your data.
Why it matters for agentic and trading-research workloads
Agent loops are input-heavy. They resend system prompts, tool definitions and accumulated context on every step, which is why the 95% cached-input discount matters more than the base rate for long-running agents. Research pipelines that summarise filings, news and on-chain data are shaped the same way: lots of reading, little writing.
The report also cites a Gray Swan prompt-injection test where Argon had a 0.7% attack success rate, against 1.0% for Opus 5.5 and 8.5% for GPT-6 Astra (lower is better). That is a vendor-adjacent benchmark, not a guarantee. If an LLM sits anywhere near order flow or credentials, you still need hard guardrails outside the model.
A practical checklist
- Re-run your own eval set and record cost per completed task, not cost per token.
- Model your monthly spend at both the introductory and the doubled rate.
- Put the model behind a thin routing layer so you can swap providers without rewriting agents.
- Keep deterministic limits (position caps, allow-lists, human approval) outside the model, whichever vendor you pick.
Bottom line
Argon is a credible, cheaper option for agent and research workloads. Treat the $2/$10 rate as a trial window and decide on measured task-level results before it ends.
FAQ
Frequently asked questions
- How much does Gemini 4 Argon cost?
- Per Cryptopolitan's report, the introductory API rate is $2 per million input tokens and $10 per million output tokens, with cached input at $0.10 per million. Those rates apply until November 30, after which pricing doubles.
- Is Gemini 4 Argon better than Claude Opus 5.5?
- Not on the independent index cited in the report. Argon scores 53 on the Artificial Analysis Intelligence Index, about five points behind Claude Opus 5.5 and Sonnet 5.5. Google's own table shows it leading in 14 of 18 results, so test it on your own tasks.
- Should I switch my trading-research agents to Argon?
- Not on price alone. Measure cost per completed task and accuracy on your own data, and model spend at the post-November-30 rate. Keep safety limits outside the model.
Sources