OpenAI releases GPT-5.4 mini and nano with 3-4x price increases but major performance gains
OpenAI has released GPT-5.4 mini and GPT-5.4 nano, compact models optimized for coding and subagent tasks. The new models deliver significant performance improvements—GPT-5.4 mini reaches 54.4% on SWE-Bench Pro versus 45.7% for GPT-5 mini—but cost 3-4x more per input token than their predecessors.
GPT-5.4 mini — Quick Specs
OpenAI Releases GPT-5.4 mini and nano: Major Performance Gains at 3-4x Price
OpenAI has launched two new compact models—GPT-5.4 mini and GPT-5.4 nano—designed for coding assistants, subagents, and computer control tasks. Both models deliver substantial capability improvements over their GPT-5 predecessors but come with steep price increases.
Pricing and Specifications
GPT-5.4 mini is available via API, Codex, and ChatGPT at $0.75 per million input tokens and $4.50 per million output tokens—a 3x increase on input and 2.25x on output versus GPT-5 mini's $0.25/$2.00 pricing.
GPT-5.4 nano is API-only at $0.20 per million input tokens and $1.25 per million output tokens—a 4x increase on input and 3.125x on output compared to GPT-5 nano's $0.05/$0.40 pricing.
Both models support a 400,000-token context window.
Performance Benchmarks
GPT-5.4 mini demonstrates substantial gains across multiple benchmark categories:
Coding Tasks:
- SWE-Bench Pro: 54.4% (vs. 45.7% for GPT-5 mini, 57.7% for GPT-5.4)
- Terminal-Bench 2.0: 60.0% (vs. 38.2% for GPT-5 mini)
Computer Control:
- OSWorld-Verified: 72.1% (vs. 42.0% for GPT-5 mini, 75.0% for GPT-5.4)
Tool Usage:
- Toolathlon: 42.9% (vs. 26.9% for GPT-5 mini)
- MCP Atlas: 57.7% (vs. 47.6% for GPT-5 mini)
Reasoning:
- GPQA Diamond: 88.0% (vs. 81.6% for GPT-5 mini, 93.0% for GPT-5.4)
GPT-5.4 nano, the smallest option, achieves 52.4% on SWE-Bench Pro and 82.8% on GPQA Diamond, making it suitable for classification, data extraction, and simpler coding subtasks.
Subagent Architecture
OpenAI showcases a multi-tiered approach in its Codex platform: GPT-5.4 handles planning, coordination, and final evaluation, while delegating parallel subtasks to GPT-5.4 mini or nano agents. These subtasks include codebase searching, large file scanning, and document processing.
According to OpenAI, this architecture allows GPT-5.4 mini to consume only 30% of GPT-5.4's quota in Codex deployments, reducing costs for simpler tasks to approximately one-third of full-model pricing.
Both models run significantly faster than their predecessors, with GPT-5.4 mini running more than twice as fast as GPT-5 mini.
What This Means
OpenAI is pursuing a deliberate strategy of performance-per-dollar at smaller model sizes rather than aggressive price competition. The 3-4x pricing increases reflect the engineering required to pack near-full-model performance into compact form factors. For developers building agentic systems and coding tools, GPT-5.4 mini's 72.1% OSWorld score represents a watershed moment—computer control capability has historically been the exclusive domain of large models.
The subagent pattern OpenAI demonstrates has immediate practical applications: orchestrating larger models for complex reasoning while routing simple tasks to cheaper compact models can reduce total API costs below full-model approaches. However, adopters should carefully benchmark against their specific use cases before upgrading from GPT-5 mini, as the pricing delta is substantial and may not justify the capability gain for all workloads.
Related Articles
Anthropic releases Fable 5, bringing capabilities of restricted Mythos model to public with $10/$50 per 1M token pricing
Anthropic has released Fable 5, making capabilities from its previously restricted Mythos model available to the public. The company claims Fable 5 beats GPT-5.5, Gemini 3.1 Pro, and its own Opus 4.8 in internal testing, with pricing set at $10 per million input tokens and $50 per million output tokens after a free trial period ending June 22.
Anthropic releases Claude Fable 5, first public Mythos-class model at $10/$50 per million tokens
Anthropic has released Claude Fable 5, its first publicly available Mythos-class model, at $10 per million input tokens and $50 per million output tokens—less than half the price of Claude Mythos Preview. The model includes safeguards that redirect sensitive queries to Claude Opus 4.8 in less than 5% of sessions.
Anthropic releases Claude Fable 5 with Mythos-class capabilities at $10/$50 per million tokens
Anthropic released Claude Fable 5, a Mythos-class model, to enterprise customers and paid subscribers two months after limiting its advanced Mythos model to select users. The new model costs $10 per million input tokens and $50 per million output tokens—twice the price of Claude Opus 4.8—and includes safeguards that block responses in high-risk areas like cybersecurity and biology.
Anthropic releases Claude Fable 5, a safety-limited version of Mythos, at $10/$50 per million tokens
Anthropic released Claude Fable 5, the first publicly available version of its Mythos model, with built-in safety restrictions that automatically block high-risk queries in cybersecurity, biology, chemistry, and related fields. The model costs $10 per million input tokens and $50 per million output tokens, double the price of Claude Opus 4.8.
Comments
Loading...