AI Model Ranking
Snapshot: 2026-08-13 · Last reviewed: 2026-08-13
Metrics follow a snapshot of the Artificial Analysis public LLM leaderboard: Intelligence Index, cost per task, and speed reflect benchmark performance at the snapshot date. Provider pricing, availability, and rankings change frequently, and a high benchmark score does not guarantee the best result for a specific codebase or workflow.
Current reasoning model
Claude Opus 5 max leads this snapshot with a top Intelligence Index of 63 and 1M context.
Best for: Frontier repository analysis, complex planning, and high-value agent tasks that need top reasoning.
Anthropic
Current reasoning model
Ties the top Intelligence Index with lower cost per task and much faster first-chunk latency than max.
Best for: Deep multi-file edits and coding agents that still need frontier quality with better latency.
Anthropic
Current reasoning model
Still among the highest Intelligence Index scores, with a 1M context window and strong output speed.
Best for: Deep repository analysis, complex planning, and high-value agent tasks where latency is acceptable.
Anthropic
Current reasoning model
High-reasoning Opus 5 mode with much lower latency than max/xhigh while staying near the top.
Best for: Daily Claude agent coding, code review, and tool-heavy workflows that still need strong reasoning.
Anthropic
Current reasoning model
Top OpenAI reasoning tier in this snapshot, with 1M context and strong Intelligence Index.
Best for: OpenAI-centered coding agents, long-context planning, and max-reasoning software tasks.
OpenAI
Current reasoning model
New Grok 4.6 high entry near the top, with 500k context and lower cost per task than the frontier Claude/OpenAI max tiers.
Best for: xAI-stack coding agents, medium-context planning, and cost-aware frontier experiments.
SpaceXAI
Current reasoning model
Moonshot's K3 max flagship with 1M+ context and strong coding scores; total response is still relatively long.
Best for: Long-horizon coding agents, frontend generation, regional API stacks, and cost-aware 1M context work.
Kimi
Current reasoning model
High-reasoning GPT-5.6 Sol mode with 1M context and lower cost per task than max.
Best for: Daily agent coding, code review, and tool-heavy workflows that still need strong reasoning.
OpenAI
Current reasoning model
Balanced Opus 5 medium tier with 1M context and much faster total response than higher tiers.
Best for: Interactive Claude coding sessions, refactors, and product engineering loops.
Anthropic
Current reasoning model
New Qwen 3.8 Max flagship with 1M context, fast first-chunk latency, and a top-10 Intelligence Index.
Best for: Multilingual coding, regional API stacks, and Qwen-centered agent experiments.
Alibaba
Current reasoning model
Balanced GPT-5.6 Sol high tier with 1M context and responsive total response time.
Best for: Interactive coding sessions, refactors, and product engineering loops inside OpenAI tools.
OpenAI
Current model; partial public metrics
Muse Spark 1.2 xhigh lands in the top 12 with 1M+ context; public speed and latency are still incomplete.
Best for: Long-context Meta model routing comparisons when Intelligence Index matters more than published latency.
Meta
Current reasoning model
GPT-5.6 Terra max tier with lower cost per task than Sol max and very fast output speed.
Best for: Cost-aware long-context coding where throughput matters more than first-token speed.
OpenAI
Current reasoning model
Grok 4.5 high with fast latency, mid-size 500k context, and lower cost per task than Grok 4.6 high.
Best for: Fast interactive coding help, medium-context agent loops, and xAI ecosystem experiments.
SpaceXAI
Current reasoning model
GPT-5.6 Sol medium tier balances intelligence, 1M context, and low total response time.
Best for: Most daily OpenAI coding loops where speed and quality both matter.
OpenAI
Current reasoning model
Sonnet-level profile with high intelligence and 1M context, trading off long first response.
Best for: Long, careful batch work where quality and context matter more than immediacy.
Anthropic
Current reasoning model
New DeepSeek V4 Pro 0813 max lands in the top 20 with 1M context, very low cost per task, and fast first-chunk latency.
Best for: Budget open-ecosystem coding agents, long-context analysis, and latency-sensitive DeepSeek Pro work.
DeepSeek
Current reasoning model
GPT-5.6 Terra xhigh combines lower cost per task, 1M context, and fast median output.
Best for: Cost-sensitive long-context coding with strong throughput and reasonable first-chunk latency.
OpenAI
Current reasoning model
High ranking with 1M context, low cost per task, and very fast first-chunk latency.
Best for: Cost-aware coding assistants, long-context analysis, and latency-sensitive workflows.
Z AI
Current reasoning model
Low-latency Opus 5 tier with 1M context and the fastest total response in the Opus 5 line.
Best for: Interactive pair programming, quick fixes, and fast planning loops on Claude Opus 5.
Anthropic
Current reasoning model
GPT-5.6 Luna max tier with very low cost per task, 1M context, and high output speed.
Best for: Budget OpenAI stacks, long-context batch edits, and high-throughput coding assistants.
OpenAI
Current reasoning model
DeepSeek V4 Flash 0731 max with 1M context, very low cost per task, and fast first-chunk latency.
Best for: Budget open-ecosystem coding agents, long-context analysis, and latency-sensitive edits.
DeepSeek
Current general model
Gemini Flash tier with 1M context, high output speed, and Google ecosystem fit.
Best for: Fast Google-stack coding loops, long-file skim, and high-throughput assistant work.
Current reasoning model
Low-latency GPT-5.6 Sol tier with 1M context and the fastest total response in the Sol line.
Best for: Interactive pair programming, quick fixes, and fast planning loops on GPT-5.6 Sol.
OpenAI
Current reasoning model
Responsive GPT-5.6 Terra high tier with 1M context and very short total response time.
Best for: Interactive OpenAI coding where throughput and low latency both matter.
OpenAI
Current reasoning model
Budget GPT-5.6 Luna xhigh tier with 1M context, very low cost per task, and fast median output.
Best for: Cost-sensitive OpenAI coding assistants and long-context batch edits.
OpenAI
Current reasoning model
Kimi K3 low tier keeps 1M+ context and lowers cost per task versus K3 max.
Best for: Cost-aware long-context Kimi coding when max reasoning effort is unnecessary.
Kimi
Current general model
Gemini 3.1 Pro preview with 1M context and a balanced cost/speed profile for Google stacks.
Best for: Google Cloud coding trials, multimodal prototypes, and long-context preview evaluation.
Current model; partial public metrics
Motif 3 lands on Intelligence Index with mid-size context; most public speed and price metrics are still incomplete.
Best for: Early evaluation of Motif 3 when benchmark rank matters more than mature public metrics.
Motif Technologies
Current reasoning model
Very low cost-per-task GPT-5.6 Luna high tier with 1M context and fast responses.
Best for: High-volume OpenAI coding assistants and budget long-context workflows.
OpenAI
Current reasoning model
Low-latency GPT-5.6 Terra medium tier with 1M context and very short end-to-end response.
Best for: Snappy interactive coding, refactors, and tool-heavy OpenAI agent loops.
OpenAI
Current model; partial public metrics
Gemini 3.5 Flash medium mode with 1M context and fast output; cost-per-task is not fully public in this snapshot.
Best for: Google ecosystem experiments where Flash-medium speed matters more than published task cost.
Current model; partial public metrics
GPT-5.3 Codex xhigh coding-focused tier; cost-per-task is not fully public in this snapshot.
Best for: Codex-centered software tasks, repo edits, and OpenAI coding agent workflows.
OpenAI
Current reasoning model
MiniMax M3 with 1M context, low cost per task, and low first-chunk latency.
Best for: Cost-aware coding assistants, regional stacks, and low-latency interactive edits.
MiniMax
Current model; partial public metrics
Motif 3 Beta lands in the ranking on Intelligence Index; most public speed and price metrics are still incomplete.
Best for: Early evaluation of emerging model providers when benchmark rank matters more than mature metrics.
Motif Technologies
Current reasoning model
Earlier DeepSeek V4 Pro max (pre-0813) with 1M context and very low cost per task; total response is longer than 0813.
Best for: Budget coding agents, self-hosted evaluation, and cost-sensitive long-context work.
DeepSeek
Current reasoning model
DeepSeek V4 Pro high tier with 1M context, low cost per task, and lower latency than max.
Best for: Cost-aware interactive coding and long-context analysis on DeepSeek endpoints.
DeepSeek
Current reasoning model
Kimi K2.7 Code coding-oriented entry with 256k context and published cost per task.
Best for: Code-focused Kimi workflows when you do not need the full 1M K3 context window.
Kimi
Current reasoning model
Xiaomi MiMo V2.5-Pro with 1M context and very low cost per task for regional stacks.
Best for: Budget multilingual coding, regional APIs, and low-cost long-context assistants.
Xiaomi
Current general model
Non-reasoning Claude Sonnet 5 with 1M context, low latency, and a lighter cost profile than max.
Best for: Everyday Claude coding, quick edits, and interactive sessions that do not need deep reasoning.
Anthropic
Current reasoning model
Inkling from Thinking Machines with 1M context, published cost per task, and solid first-chunk latency.
Best for: Long-context evaluation of newer model labs with published pricing and latency.
Thinking Machines
Current reasoning model
Tencent Hy3 with mid-size context, very low cost per task, and competitive first-chunk latency.
Best for: Regional coding stacks, budget assistants, and Tencent-ecosystem experiments.
Tencent
Current model; partial public metrics
Nex N2-Pro enters with mid-size context and fast median output; cost-per-task is not fully public here.
Best for: Emerging-provider evaluation and low-latency coding experiments outside the major labs.
Nex AGI
Current general model
Non-reasoning GPT-5.6 Sol with 1M context and the lowest first-chunk latency in the Sol line.
Best for: Snappy OpenAI coding loops when extended reasoning is unnecessary.
OpenAI
Current model; partial public metrics
Upstage Solar Pro 4 with 512k context and published speed; cost-per-task is not fully public in this snapshot.
Best for: Regional and document-heavy coding stacks evaluating Upstage models on Intelligence Index.
Upstage
Current reasoning model
Low-latency GPT-5.6 Terra low tier with 1M context and very short end-to-end response.
Best for: Fast interactive OpenAI coding and quick refactors where max reasoning is unnecessary.
OpenAI
Current reasoning model
Smaller Inkling variant with 1M context, low cost per task, and faster median output than base Inkling.
Best for: Budget evaluation of Thinking Machines models with tighter latency and cost targets.
Thinking Machines
Current model; partial public metrics
China Mobile JT-4.1 Flash lands on Intelligence Index; most public price and speed metrics remain incomplete.
Best for: Regional China Mobile stack evaluation when benchmark rank is the primary signal.
China Mobile
Current model; partial public metrics
Agnes 2.5 Pro Alpha with 1M context and fast median output; cost-per-task is not fully public here.
Best for: Early long-context evaluation of emerging Sapiens AI models.
Sapiens AI
Current reasoning model
Qwen 3.7 Plus with 1M context, lower cost per task than Qwen 3.8 Max, and moderate total response time.
Best for: Cost-aware Qwen coding agents and multilingual product engineering loops.
Alibaba