AIToolCrunch
← For Developers
AI Models

AI Models Compared

Every major frontier model - pricing, context window, benchmark scores, and what each one is actually best at. Continuously web-verified; last checked 2026-07-05.

Pricing at a glance

ModelProviderStrong atContextInput / 1MOutput / 1MType
DeepSeek V4 FlashPreviewDeepSeekLong context1M tokensFreeFreeOpen
DeepSeek V4PreviewDeepSeekLong context1M tokensFreeFreeOpen
GPT-5.5OpenAICoding1,050,000 tokens (128,000 max output)$5.00$30.00Closed
Claude Opus 4.8AnthropicCoding1M$5.00$25.00Closed
Claude Sonnet 4.6AnthropicLong context1M$3.00$15.00Closed
Claude Opus 4.7AnthropicCoding1M$5.00$25.00Closed
Claude Haiku 4.5AnthropicSpeed200K$1.00$5.00Closed
GPT-5OpenAIMath400,000 tokens (128,000 max output)$1.25$10.00Closed
GPT-4oOpenAISpeed128,000 tokens (16,384 max output)$2.50$10.00Closed
o1OpenAIReasoning200,000 tokens (100,000 max output)$15.00$60.00Closed
Gemini 3.5Google DeepMindCoding1,048,576 tokens (Gemini 3.5 Flash; Pro variant not yet released)$1.50$9.00Closed
Gemini 2.5 ProGoogle DeepMindLong context1,048,576 tokens (1M) input; up to 65K output$1.25$10.00Closed
Gemini 2.5 FlashGoogle DeepMindCost efficiency1,048,576 tokens (1M) input; up to 65,535 output$0.30$2.50Closed
Amazon Nova ProAmazon Web ServicesMultimodal300K tokens$0.80$3.20Closed
DeepSeek V3DeepSeekCost efficiency128K tokensFreeFreeOpen
Llama 4MetaLong contextUp to 10M tokens (Scout); ~1M tokens (Maverick)FreeFreeOpen
Qwen 3Alibaba (Qwen Team)Cost efficiency128K tokens (32K for 0.6B/1.7B/4B dense variants)FreeFreeOpen
Mistral LargeMistral AIReasoning1280002.006.00Closed
Gemma 3Google DeepMindCost efficiency128K tokens (32K for the 1B variant)FreeFreeOpen
Command R+CohereLong context128K tokens$2.50$10.00Closed
Gemma 4 12BGoogleCost efficiency256K tokensFreeFreeOpen
Claude Fable 5AnthropicCoding1M$10.00$50.00Closed
North Mini CodeCohereCost efficiency256K tokens00Closed
GPT-5.4OpenAILong context1,050,000 tokens (128,000 max output)$2.50$15.00Closed
GPT-5.5 InstantOpenAISpeed10500005.0030.00Closed
Mistral OCR 4Mistral AICost efficiencyNot announced40004000Closed
GPT-5.6PreviewOpenAICodingNot announced530Closed
GPT-5.6 SolPreviewOpenAICodingNot announced$5.00$30.00Closed
Claude Sonnet 5AnthropicCoding1M$3.00$15.00Closed
FlintPreviewSpringboardsSpeedNot announcedNot announcedNot announcedClosed
Nano Banana 2 LiteGoogleSpeedNot announcedNot announced33.6Closed

Prices in USD. Open-source models are free to self-host; API pricing varies by provider.

Closed-source models

GPT-5.5

1,050,000 tokens (128,000 max output) ctx

OpenAI

OpenAI's smartest general-purpose frontier model for professional work

Agentic coding and multi-tool automationLarge-context document and codebase analysisGeneral-purpose frontier reasoning and knowledge work
Full details →

Claude Opus 4.8

1M ctx

Anthropic

Anthropic's top general-availability workhorse for complex agentic coding and enterprise work.

Complex, long-horizon agentic coding and autonomous engineeringEnterprise knowledge work and professional-grade analysisHigh-stakes tasks where honesty and flaw-catching matter
Full details →

Claude Sonnet 4.6

1M ctx

Anthropic

Anthropic's most capable Sonnet-class model of early 2026, now superseded by Sonnet 5.

Established mid-tier production workloads pinned to a stable modelLarge-context document and codebase analysisAgentic coding at a moderate price point
Full details →

Claude Opus 4.7

1M ctx

Anthropic

The April 2026 Opus flagship - top-tier coding and vision, now superseded by Opus 4.8.

Vision-heavy agentic and document-understanding workloadsTop-tier coding on established Opus 4.7 pipelinesLarge-codebase engineering with 1M context
Full details →

Claude Haiku 4.5

200K ctx

Anthropic

Anthropic's fastest, cheapest model with near-frontier intelligence.

High-throughput, low-latency production workloadsParallelized sub-agents and multi-agent worker rolesCheap classification, extraction, and summarization at scale
Full details →

GPT-5

400,000 tokens (128,000 max output) ctx

OpenAI

OpenAI's landmark August 2025 flagship: strong reasoning at a low price

Budget-friendly frontier reasoning and codingMath and STEM problem solvingExisting GPT-5-based production systems
Full details →

GPT-4o

128,000 tokens (16,384 max output) ctx

OpenAI

OpenAI's versatile, fast multimodal workhorse (text + image)

Fast, low-cost everyday assistant tasksMultimodal (image + text) understandingHigh-volume production workloads
Full details →

o1

200,000 tokens (100,000 max output) ctx

OpenAI

OpenAI's first-generation deep-reasoning model that thinks before answering

Hard math and science reasoningCompetitive and algorithmic programmingDeliberate multi-step problem solving
Full details →

Gemini 3.5

1,048,576 tokens (Gemini 3.5 Flash; Pro variant not yet released) ctx

Google DeepMind

Google's frontier model for agents and coding, made fast and cheap.

Autonomous coding agentsHigh-volume production apps needing frontier qualityMultimodal document and chart understanding
Full details →

Gemini 2.5 Pro

1,048,576 tokens (1M) input; up to 65K output ctx

Google DeepMind

Google's advanced thinking model for complex reasoning, coding, and long context.

Complex reasoning and STEM problem-solvingLong-context document and codebase analysisHigh-quality multimodal understanding
Full details →

Gemini 2.5 Flash

1,048,576 tokens (1M) input; up to 65,535 output ctx

Google DeepMind

Google's price-performance workhorse with thinking and a 1M-token context.

High-volume production workloadsCost-sensitive chat and RAG appsFast multimodal processing
Full details →

Amazon Nova Pro

300K tokens ctx

Amazon Web Services

Amazon's balanced multimodal Bedrock model for text, image, and video at scale.

AWS/Bedrock-native applicationsMultimodal text/image/video tasksCost-balanced general workloads
Full details →

Mistral Large

128000 ctx

Mistral AI

Mistral's state-of-the-art, open-weight, general-purpose multimodal flagship.

Multilingual reasoning and generationStructured/JSON output and codingOpen-weight flagship deployments
Full details →

Command R+

128K tokens ctx

Cohere

Cohere's RAG- and tool-use-optimized model, still live but superseded by Command A.

Retrieval-augmented generation with citationsMulti-step tool-use / agentsMultilingual enterprise assistants
Full details →

Claude Fable 5

1M ctx

Anthropic

Anthropic's most capable widely released model - frontier intelligence for long-running agents.

Long-running autonomous agents on complex, high-value tasksFrontier software engineering and hard multi-service implementationScientific research and advanced analytics
Full details →

North Mini Code

256K tokens ctx

Cohere

Cohere's first open-weight agentic coding model - 30B MoE, 3B active, runs on one H100.

Agentic software engineeringTerminal/tool-driven coding agentsSelf-hosted sovereign coding on one GPU
Full details →

GPT-5.4

1,050,000 tokens (128,000 max output) ctx

OpenAI

Capable, cost-efficient predecessor to GPT-5.5 with a 1M+ context window

Cost-efficient large-context workGeneral coding and reasoning at scaleProduction workloads migrating within the GPT-5 generation
Full details →

GPT-5.5 Instant

1050000 ctx

OpenAI

The fast, default ChatGPT model tuned for low-latency responses

Fast, everyday ChatGPT interactionsLatency-sensitive assistant and chat useQuick drafting and summarization
Full details →

Mistral OCR 4

Not announced ctx

Mistral AI

State-of-the-art document OCR that turns PDFs and scans into structured Markdown.

Document OCR and parsingPDF/scan to structured Markdown or JSONMultilingual document intelligence
Full details →

GPT-5.6

Not announced ctx

OpenAI

OpenAI's next-generation GPT-5.6 model family: Sol, Terra, and Luna

Teams that want one generation with multiple cost/quality tiersAgentic coding, knowledge work, and research at scaleEarly adopters and approved preview partners
Full details →

GPT-5.6 Sol

Not announced ctx

OpenAI

OpenAI's most capable and security-hardened frontier model, in limited preview

Frontier agentic coding and computer useComplex multi-step tasks via subagent orchestrationSecurity-sensitive and scientific research workloads
Full details →

Claude Sonnet 5

1M ctx

Anthropic

The best combination of speed and intelligence, at near-Opus quality for a Sonnet price.

Agentic coding assistants and autonomous dev agentsHigh-volume production agent loops needing strong reasoning per dollarLarge-codebase and long-document analysis with 1M context
Full details →

Flint

Not announced ctx

Springboards

A divergence model designed to generate diverse, creative outputs instead of converging on predictable answers

Advertising and marketing ideationCreative brainstormingStrategic exploration
Full details →

Nano Banana 2 Lite

Not announced ctx

Google

The fastest, most cost-efficient image generation model in the Nano Banana family

High-volume image generationRapid ideation and A/B testingIterative content creation
Full details →

Open-source models

Free to download, self-host, and fine-tune.

DeepSeek V4 Flash

Open

DeepSeek

Compact 284B-param open MoE that keeps a 1M context at a fraction of Pro's cost.

High-volume, cost-sensitive inferenceLong-context tasks on a budgetSelf-hosting on modest hardware
Full details →

DeepSeek V4

Open

DeepSeek

Open-weight 1.6T-param MoE frontier model with a 1M-token context built for agents.

Self-hosted frontier reasoning and codingLong-context agentic workflowsCost-sensitive high-volume inference
Full details →

DeepSeek V3

Open

DeepSeek

The open-weight 671B-param MoE that put DeepSeek on the frontier map.

Cost-efficient open-weight general useSelf-hosting and fine-tuningCoding and instruction-following
Full details →

Llama 4

Open

Meta

Meta's natively multimodal open MoE herd with industry-leading context length.

Extremely long-context open-model applicationsSelf-hosted multimodal deploymentsEfficient MoE inference
Full details →

Qwen 3

Open

Alibaba (Qwen Team)

Alibaba's open-weight model family with switchable thinking and non-thinking modes.

Self-hosted reasoning and coding assistantsApplications needing switchable reasoning depthMultilingual open-model deployments
Full details →

Gemma 3

Open

Google DeepMind

Google's open, multimodal, multilingual long-context model family.

Local and on-prem multimodal assistantsMultilingual applicationsFine-tuning on custom data
Full details →

Gemma 4 12B

Open

Google

Google's laptop-runnable open multimodal model with a unified encoder-free design.

Self-hosted multimodal assistantsOn-device / single-GPU deploymentPrivacy-sensitive applications
Full details →

Benchmark scores sourced from official provider release posts. Prices subject to change - check provider pricing pages for current rates.