Gemini 4 Neon Leaked: Intelligence Index Score of 61 Rivals Opus/Sol

Google's mid-tier Gemini 4 Neon scores 61 on the Intelligence Index, rivaling flagships, as Anthropic sets its IPO at $83.
According to RSS reports, Google's internally codenamed mid-tier model Gemini 4 Neon has surfaced with an Intelligence Index score of 61, seen as a direct competitor to Anthropic's Opus and Sol series, suggesting Google is pushing flagship capabilities down to mid-range pricing. The same briefing disclosed Anthropic's IPO price of $83, with both leading vendors releasing product and capital signals almost simultaneously. The article notes that mid-tier models carry the heaviest commercialization burden, and a score of 61 means developers may get near-flagship reasoning in cost-sensitive scenarios; but a single index can't cover multimodality, long context, or tool-calling stability, so hands-on testing is still needed. The overall judgment is that AI competition has shifted from leaderboard battles to a two-front game of commercialization efficiency and product matrices, while current information comes from a single source with parameters and release timing undisclosed.
Gemini 4 Neon Surfaces: Google's Mid-Tier Model Aims Above Its Weight
According to RSS channel reports, Google's internally codenamed model Gemini 4 Neon has entered public view for the first time. Positioned as a "mid-tier" model, it scored 61 on the Intelligence Index and is widely seen as a direct competitor to Anthropic's Opus and Sol series. The signal here is subtle: Google appears to be redrawing the boundaries of its own model hierarchy, pushing capabilities that once belonged to the flagship tier down to a mid-range price point.
Judging by naming logic, "Neon" continues Google's habit of distinguishing model variants with color and material codenames, while "Gemini 4" indicates it belongs to the fourth-generation product family. Notably, this news appeared in the same briefing as Anthropic's IPO pricing — the two companies released product and capital signals almost simultaneously, and competition among leading AI vendors is shifting from pure leaderboard battles to a two-front game spanning product matrices and capital markets.
Why the Mid-Tier Positioning Matters
Across the product lines of major LLM vendors, mid-tier models typically carry the heaviest commercialization burden: they need to be cheap enough to support large-scale calls, yet smart enough to cover real business scenarios. Gemini 4 Neon's Intelligence Index score of 61 suggests Google is trying to deliver near-flagship reasoning performance at this price point. For developers, this could mean that in cost-sensitive applications, they may no longer be forced to choose between "cheap but dumb" and "smart but expensive."
However, a single index score cannot fully capture a model's capabilities. The Intelligence Index usually emphasizes reasoning and knowledge tasks, with limited coverage of multimodal ability, long context, and tool-calling stability. Whether Gemini 4 Neon continues Google's traditional strength in multimodality remains unknown, as no public information is available yet.
The Intelligence Index is a type of comprehensive evaluation benchmark, usually maintained by third-party organizations or communities. It aggregates dozens of standardized tests (such as mathematical reasoning, science Q&A, code generation, and common-sense judgment) into a single score for easy cross-model comparison. Its advantage is being intuitive and rankable, but its limitations are equally obvious: first, the test set itself may be covered by model training data, inflating scores; second, it struggles to reflect a model's robustness in real business settings, such as long-document understanding, multi-turn conversation consistency, and function-calling format accuracy. Therefore, 61 should be understood as a "relative position within a specific evaluation distribution" rather than an absolute capability scale. A common industry practice is to cross-reference it with human preference ratings from LMSYS Chatbot Arena and vendors' self-reported benchmark results to form a more complete judgment.
Anthropic's IPO Pricing: The Capital Narrative Behind $83
The same report disclosed that Anthropic has set its IPO price at $83. This figure itself does not constitute a judgment on model capability, but it reveals a key fact: leading AI labs are accelerating their entry into public capital markets. For AI companies that have long relied on private financing, an IPO means stricter financial disclosure and more direct valuation pressure, and it also means they must prove to investors that model capabilities can translate into sustained revenue.
Placing the two pieces of information side by side: Google is expanding product coverage with mid-tier models, while Anthropic is anchoring capital expectations with an IPO — both paths point to the same conclusion: AI competition has entered a phase of competing on commercialization efficiency, and the marginal value of simply refreshing leaderboard scores is declining.
Practical Implications for Developers
For teams currently choosing models, the emergence of Gemini 4 Neon provides a new price-performance reference point. If the Intelligence Index score of 61 can be reliably reproduced in actual API calls, it would be attractive for agent workflows, batch content processing, code assistance, and similar scenarios. But selection decisions still require hands-on testing: the same score can vary widely across different task distributions, especially in scenarios involving structured output and multi-turn tool calling.
AI lab IPOs differ structurally from those of traditional software companies. Because training costs run into hundreds of millions of dollars and require sustained investment, post-IPO companies must balance "maintaining R&D intensity" with "delivering profits to public markets." Quarterly earnings reports will directly expose previously opaque metrics such as inference costs, gross margins, and customer concentration. The $83 pricing is essentially a collective discount by investment banks and institutional investors of the company's future revenue curve — it includes both a premium for model capabilities and a discount for compute supply chain, regulatory risk, and competitive landscape. For downstream developers, one observable consequence of an IPO is that vendors, to support valuations, tend to more actively roll out tiered pricing, enterprise SLAs, and volume commitment discounts, further refining the mid-tier segment of their product matrices.
Agent workflow refers to a system architecture in which a large model autonomously plans steps, calls external tools (search, code execution, database queries, etc.), and iterates over multiple rounds to complete a task. The requirements such scenarios place on models differ significantly from single-turn Q&A: they need stable structured output (such as JSON mode), fault-tolerant parsing of tool return results, and retention of the original goal across multi-turn contexts. Models with high evaluation scores may not perform well on these dimensions, because most public benchmarks do not measure tool-calling format compliance rates or long-chain task completion rates. When selecting models, it is advisable to construct a "golden test set" close to your own business, fix prompts and tool definitions, and run A/B tests on candidate models, focusing on failure modes rather than average scores.
Short-Term Highlights of the Competitive Landscape
This round of moves by Google and Anthropic puts the Opus and Sol series under more direct price-performance comparison. Intensified competition in the mid-tier model market typically translates into API pricing adjustments and changes to free quotas within weeks to months. For developers, continuously tracking each vendor's model matrix is more worthwhile than locking into a single provider.
It should be noted that the current information comes from a single RSS briefing, and Gemini 4 Neon's specific parameters, context window, and release timeline have not been disclosed. The Intelligence Index score of 61 should be treated as a preliminary signal rather than a final verdict. Only if multiple independent evaluations cross-validate in the future can a reliable judgment be made about the model's true level.
Related articles

Gemini 4 Neon's Comeback: How Google's 'American Doubao' Shipped Five Models in Two Weeks
Google shipped five models in two weeks. Gemini 4 Neon ranks third globally and first on the coding agent leaderboard, at half the cost of Opus 5.5. A Bilibili creator breaks down the three-layer comeback story: regulatory relief, sparse MoE architecture, and in-house TPUs—plus Carbon promo savings tips and AntiGravity quota rules.

What Is Hermes Agent? An Always-On AI Agent That Grows With You
Hermes Agent is an always-on AI agent running on a VPS, reachable via WhatsApp, Telegram, and Discord. It manages lists, calendars, and scheduled tasks and grows with use.

Building AI Animated Videos with Claude Code: A Hands-On Local Toolchain Guide
A hands-on demo of using Claude Code with local open-source tools like Kokoro and Remotion to generate AI videos with narration, animation, and music in minutes. Workflow, efficiency gains, and limitations analyzed.