Claude Opus 4.8, GPT-5.5, Gemini 3.5 Flash, Grok 4.3, and DeepSeek V4: Full API Pricing Table

Choosing the right model in 2026 isn't just about benchmark scores—it's about understanding the real cost of tokens. With Anthropic, OpenAI, Google, xAI, and DeepSeek all shipping fresh flagships, the per-token price tags have become a primary battleground. This comparison breaks down the latest verified API pricing for Claude Opus 4.8, GPT-5.5, Gemini 3.5 Flash, Grok 4.3, and DeepSeek V4, including how caching, context length, and tokenizer quirks change the bill.

Head-to-Head Standard API Pricing

All prices are per 1 million tokens (MTok) in USD, based on official documentation and verified third-party sources as of late May 2026.

Model	Provider	Standard Input $/MTok	Cached Input $/MTok	Standard Output $/MTok	Context Window	Max Output
Claude Opus 4.8	Anthropic	$5.00	$0.50	$25.00	1M tokens	128K tokens
Claude Opus 4.7	Anthropic	$5.00	$0.50	$25.00	1M tokens	128K tokens
GPT-5.5	OpenAI	$5.00	$0.50	$30.00	Short-context tier; long-context surcharge above ~272K tokens	128K tokens
GPT-5.5 Pro	OpenAI	$30.00	—	$180.00	Short-context tier; long-context surcharge above ~272K tokens	128K tokens
Gemini 3.5 Flash	Google	$1.50	$0.15	$9.00	1,048,576 tokens	65,536 tokens
Grok 4.3	xAI	$1.25	$0.20	$2.50	1M tokens	—
DeepSeek V4 Flash	DeepSeek	$0.14	$0.028	$0.28	1M tokens	384K tokens
DeepSeek V4 Pro	DeepSeek	$1.74	$0.145	$3.48	1M tokens	384K tokens

Claude Opus 4.8 and 4.7 share the same standard rates as Opus 4.6, continuing Anthropic's pattern since the Opus 4.5 generation . However, Opus 4.7 introduced a new tokenizer that can produce up to 35% more tokens for the same input text compared to Opus 4.6, which effectively raises the cost of identical prompts even though the per-token price is unchanged . GPT-5.5's pricing, meanwhile, doubles for input and increases 1.5x for output once a request exceeds roughly 272K tokens, a tiered structure that can surprise users who aren't monitoring prompt length . Grok 4.3 applies a similar approach: standard rates apply for requests up to 200K tokens, after which prices double .

DeepSeek V4 Flash stands apart as the least expensive model in this group by a wide margin. At $0.14 input and $0.28 output per million tokens, it is roughly 97% cheaper than GPT-5.5 on output and nearly 70% cheaper than Grok 4.3 on input, making it a leading choice for high-volume agentic workloads .

Caching Discounts: How Much You Can Save

Prompt caching is the single most effective way to lower per-request costs across all providers. When prompts share a common prefix—like system instructions or long conversation histories—cached input tokens are billed at a fraction of the standard rate.

Model	Cached Input $/MTok	Discount vs. Standard Input
Claude Opus 4.8 / 4.7	$0.50	~90% off
GPT-5.5 (short context)	$0.50	90% off
GPT-5.5 (long context)	$1.00	90% off
Gemini 3.5 Flash	$0.15	90% off
Grok 4.3	$0.20	~84% off
DeepSeek V4 Flash	$0.028	80% off
DeepSeek V4 Pro	$0.145	~92% off

Anthropic, OpenAI, and Google all converge on a roughly 90% discount for cached input. DeepSeek lists its V4 Flash cache-hit rate at $0.028, which is an 80% reduction from the $0.14 cache-miss price . Grok 4.3's cached rate was introduced at $0.20 per million tokens for requests under 200K tokens . For workloads with repetitive prompts, these caching tiers can easily cut monthly API bills by half or more.

Batch Processing Costs

Batch APIs offer another major cost lever, generally halving standard per-token prices in exchange for slower turnaround times.

• Anthropic Claude Opus 4.7/4.8: Batch input is $2.50/MTok and batch output is $12.50/MTok, exactly 50% off the standard rates, with the same discount applying to Sonnet and Haiku tiers .
• OpenAI GPT-5.5: Batch and Flex pricing are listed at half the standard API rate—$2.50 input and $15.00 output per MTok .
• DeepSeek V4: The official DeepSeek pricing page shows an additional 75%-off discount on V4 Pro pricing from launch rates, bringing it to $0.435 cache-miss input and $0.87 output on some providers, though batch-specific rates weren't confirmed in the provided sources .

Context Windows and Output Limits

Context window size and maximum output tokens influence both capability and cost. A larger context window means more input tokens per request, which directly multiplies the bill.

• Claude Opus 4.8 & 4.7 support 1M context tokens and up to 128K output tokens with no surcharge for long-context requests .
• GPT-5.5 supports a 1.05M–1.1M context window but applies a steep long-context pricing multiplier once input exceeds roughly 272K tokens .
• Gemini 3.5 Flash offers 1,048,576 input tokens and 65,536 output tokens .
• Grok 4.3 provides a 1M context window but doubles rates above 200K total tokens .
• DeepSeek V4 Flash & Pro share a 1M context window and 384K maximum output, the largest output window in this group .

Key Takeaways

• Best budget frontier model: Grok 4.3 ($1.25/$2.50) for most use cases, with DeepSeek V4 Flash ($0.14/$0.28) as the cheapest option for cost-sensitive workloads .
• Most expensive: GPT-5.5 Pro ($30/$180), aimed at high-accuracy enterprise workloads .
• Claude Opus 4.7 hidden cost: The 35% tokenizer increase means identical prompts can cost more than on Opus 4.6, despite an unchanged per-token price .
• Caching is essential: Across all providers, prompt caching reduces input costs by 80–92%, making it the first optimization step for any production deployment .
• Watch the long-context multipliers: Both OpenAI and xAI double or 1.5x prices for large prompts, while Anthropic and DeepSeek keep flat pricing across the full context window .
• Open-source options: DeepSeek's V4 series can be self-hosted for even lower effective costs, with providers like OpenRouter listing V4 Pro at $0.435/$0.87 .