Agnes AI Launches Agnes 2.5 Pro Alpha: A Budget Reasoning Model That Punches Above Its Price

Singapore's Agnes AI has entered the crowded reasoning-model race with **Agnes 2.5 Pro Alpha**, a new proprietary model that pairs mid-tier intelligence with aggressively low pricing. According to Artificial Analysis, the model debuts at 39 on the Artificial Analysis Intelligence Index while undercutting nearly every rival on cost.

## Who Is Agnes AI?

Agnes AI is a Singapore-based AI lab that trains its own full-modality foundation models in-house across text, image, and video, and offers them through a free omni-modal API that has passed 3 million users. The company has been building momentum over the past several months. Back in June 2026, Agnes AI unveiled milestones including becoming the first Singapore-born AI model company to be listed on the Artificial Analysis global benchmark leaderboard with a top 10 AI lab placement, opening free access to its complete model suite for a limited period, and joining Singapore's National AI Impact Program in partnership with IMDA and AI Singapore to support upskilling 40,000 technology professionals.

That leaderboard presence spans multiple modalities. Agnes AI reached global top 10 AI lab placements across three distinct model categories: Agnes-2.0-Flash for text and agentic capability on Claw-Eval, Agnes-Image-2.0-Flash for image editing, and Agnes-Video-V2.0 for video generation on Artificial Analysis. Notably, the standings position Agnes AI alongside AI laboratories from Anthropic, OpenAI, Google, KlingAI, ByteDance, and Alibaba.

The company also positions itself as a challenger on accessibility. Its consumer-facing platform, sometimes associated with developer "Sapiens AI," stands out in a market where most AI APIs charge per token by offering free API credits, free API keys, and access to state-of-the-art models, alongside a broader product suite that includes an all-in-one assistant, a companionship-focused product, and a generative content platform.

## What Is Agnes 2.5 Pro Alpha?

Agnes 2.5 Pro Alpha is a text, image, and video input reasoning model with text output. Per Agnes AI's own documentation, it is a paid reasoning model from Agnes AI designed for advanced coding, scientific reasoning, long-context analysis, agentic workflows, and multimodal understanding. The company recommends it specifically for tasks where correctness and multi-step reasoning matter more than raw latency, such as scientific reasoning, complex policy analysis, and long-form technical planning.

Artificial Analysis lists the model as a proprietary reasoning model released on July 24, 2026. It's accessed via the OpenAI-compatible Chat Completions API, using the model identifier `agnes-2.5-pro-alpha`.

### Technical Specs

- **Context window:** 1 million tokens - **Input modalities:** Text, image, and video - **Output modality:** Text only - **Pricing:** $0.45 per 1M input tokens / $0.90 per 1M output tokens / $0.0038 per 1M cache-hit tokens - **Availability:** Agnes AI first-party API (currently one benchmarked provider, "Agnes NG")

## Benchmark Performance

Artificial Analysis's independent testing found that Agnes 2.5 Pro Alpha scores 39 on the Artificial Analysis Intelligence Index, placing it mid-pack among reasoning models tested and below the current proprietary frontier. That places it well ahead of the typical model in its price bracket, though — the model's own product page notes it places well above average among other reasoning models in a similar price tier, where the median score is just 16.

**Coding** stands out as a particular strength. The model scores 58.8 on the Coding Index, near the top for its intelligence tier — above similarly priced models including DeepSeek V4 Flash (56.2), GPT-5.4 mini (56.1) and Qwen3.7 Plus (55.9), and just behind Nex-N2-Pro (59.1).

**Academic reasoning** is also competitive. Agnes 2.5 Pro Alpha posts 32% on Humanity's Last Exam and 88% on GPQA Diamond, with the Humanity's Last Exam result leading similarly priced proprietary models including GPT-5.4 mini (xhigh, 27%) and GPT-5.4 nano (xhigh, 26%) according to the original announcement text, while its GPQA Diamond score lines up with peers at a similar Intelligence Index level.

Interestingly, the model achieves these results with relatively modest token usage. It used 22k output tokens per Artificial Analysis Intelligence Index task, 16k of them reasoning tokens — 51% fewer than DeepSeek V4 Flash (max, 45k) and 72% fewer than GPT-5.4 mini (xhigh, 78k).

On speed, benchmarking shows the model generating output at 99.2 tokens per second, above average compared to other reasoning models in a similar price tier (median: 96.9 t/s), with a time to first token of 1.72 seconds, better than the tier median. Provider-level testing recorded an output speed of 96.7 t/s and a time to first answer token of 22.30s through the Agnes NG provider.

## Pricing: The Real Headline

Price is where Agnes 2.5 Pro Alpha makes its strongest case. Analysts note that at $0.45/$0.90 per 1M input/output tokens, it is the second cheapest model at its intelligence level, and its defining characteristic is price — among the lowest-priced proprietary models at its intelligence tier.

Using Artificial Analysis's standard blended pricing methodology (a 7:2:1 cache-hit/input/output ratio), the model costs just $0.18 per 1M tokens — behind only DeepSeek V4 Flash (Reasoning, Max Effort) among models scoring 38–40 on the Intelligence Index. Open-weight alternatives like DeepSeek V4 Pro (Reasoning, Max Effort) and MiMo-V2.5-Pro match that $0.18 price point but at higher measured intelligence.

Head-to-head comparisons reinforce the value proposition: - Versus **GPT-5.2 (xhigh)**: Agnes 2.5 Pro Alpha is cheaper, costing $0.18 per 1M tokens compared with GPT-5.2's pricing, and also faster, generating 131.7 tokens per second versus GPT-5.2's 71.3 tokens per second — though GPT-5.2 (xhigh) is more intelligent, scoring 42 (estimated) versus Agnes's 39. - Versus **MiniMax-M2.1**: Agnes 2.5 Pro Alpha is more intelligent, scoring 39 compared with MiniMax-M2.1's estimated 31, while also being cheaper at $0.18 per 1M tokens versus MiniMax-M2.1's $0.39. - Versus **GLM-5.1 (Non-reasoning)**: Agnes comes out ahead on both intelligence and price, scoring 39 versus GLM-5.1's estimated 35, at $0.18 per 1M tokens versus $0.90.

## How to Access It

Developers can reach the model through Agnes AI's OpenAI-compatible Chat Completions API. Documentation notes that users should use `agnes-2.5-pro-alpha` as the model name and confirm their account has access to the paid model, with basic requests requiring only a model and messages field, and image inputs passed via publicly accessible URLs.

## The Bigger Picture

Agnes 2.5 Pro Alpha arrives as Agnes AI continues stacking up benchmark credentials across modalities — following prior top-10 placements for its Flash-series text, image, and video models — and as the company pushes into national initiatives like Singapore's AI upskilling program. While it doesn't challenge the current proprietary frontier (currently topped by models like Claude Opus 5), Agnes 2.5 Pro Alpha's combination of solid coding performance, competitive academic reasoning scores, and rock-bottom pricing makes it one of the more compelling budget options for developers prioritizing cost-efficiency over top-end intelligence.