Standard Compute
Live Model Rankings 2026

Best AI Model for Agents 2026

Which model should you run in your agent? Compare the models people actually use — Claude, GPT-5, Gemini, DeepSeek, Qwen, GLM, Kimi and more — ranked by output quality, agentic ability, speed, reliability, and value for money, with live pricing from OpenRouter. 29,397 community ratings so far — add yours.

Live Ranking

The Best AI Models for Agents, Ranked

Overall ranks by capability for agents — weighted toward agentic ability and output quality, calibrated against public benchmarks (SWE-bench, Terminal-Bench, Artificial Analysis, LMArena). Price isn't in this score: switch to Value for $ to rank by quality per dollar instead. Tap any model for the full breakdown.

1
Claude Fable 5Anthropic · 1M ctx · $$$$
Anthropic's Mythos-class flagship (June 2026) — the new benchmark leader for agentic coding, at a premium price.
97/100
$22/1M · 97 community
2
GPT-5.6 SolOpenAI · 1.1M ctx · $$$$
OpenAI's new flagship tier (July 2026) — built for the hardest coding and research problems.
95/100
$13/1M · 95 community
3
Claude Opus 4.8Anthropic · 1M ctx · $$$$
The output-quality benchmark for agentic coding — the model most coding agents reach for on hard tasks.
94/100
$11/1M · 94 community
4
GPT-5.5OpenAI · 1.1M ctx · $$$$
OpenAI's flagship — frontier reasoning and the model behind Codex's hardest runs.
93/100
$13/1M · 93 community
5
Claude Sonnet 4.6Anthropic · 1M ctx · $$$$
The default daily driver for agents — most of Opus's quality at a fraction of the price.
91/100
$6.60/1M · 91 community
6
GPT-5.6 TerraOpenAI · 1.1M ctx · $$$
The balanced tier of the GPT-5.6 family — GPT-5.5-class performance at half the price.
90/100
$3.13/1M · 90 community
7
GPT-5.4OpenAI · 1.1M ctx · $$$$
The value flagship of the GPT-5 line — most of 5.5's ability for less than half the price.
90/100
$6.25/1M · 90 community
8
Gemini 3.1 ProGoogle · 1.0M ctx · $$$
Google's flagship — frontier quality with a massive multimodal 1M-token context.
89/100
$5.00/1M · 89 community
9
GLM 5.2Z.AI · 1.0M ctx · $$
Z.AI's flagship open-weight model — strong agentic coding with a 1M context.
89/100
$1.26/1M · 89 community
10
Gemini 3.5 FlashGoogle · 1.0M ctx · $$$
Fast, capable, and cheap — a favourite default for high-throughput agents.
89/100
$3.75/1M · 89 community
11
Qwen3.7 MaxAlibaba (Qwen) · 1M ctx · $$$
Qwen's top hosted model — frontier-class quality with a 1M context at a fair price.
89/100
$2.36/1M · 89 community
12
GPT-5.3-CodexOpenAI · 400K ctx · $$$
OpenAI's coding-tuned model — the default behind Codex CLI's local runs.
89/100
$5.42/1M · 89 community
13
Kimi K2.7 CodeMoonshot AI · 262K ctx · $$$
Open-weight model tuned for agentic coding — a community favourite for cheap, capable dev work.
88/100
$1.56/1M · 88 community
14
DeepSeek V4 ProDeepSeek · 1.0M ctx · $$
Open-weight near-frontier quality at roughly a tenth of flagship pricing.
88/100
$0.57/1M · 88 community
15
Kimi K2.6Moonshot AI · 262K ctx · $$
Moonshot's flagship open-weight generalist — broad capability with a long context.
87/100
$1.27/1M · 87 community
16
GPT-5.6 LunaOpenAI · 1.1M ctx · $$
The fast, cheap tier of the GPT-5.6 family — routine work and subagents at $1/$6 per million.
85/100
$1.25/1M · 85 community
17
MiniMax M3MiniMax · 1.0M ctx · $$
An open-weight model built for agents — strong tool use with a 1M context at low cost.
85/100
$0.57/1M · 85 community
18
Claude Haiku 4.5Anthropic · 200K ctx · $$$
Anthropic's fast, cheap model for high-volume agent steps and subagents.
84/100
$2.20/1M · 84 community
19
Qwen3.7 PlusAlibaba (Qwen) · 1M ctx · $$
The value workhorse of the Qwen line — cheap, fast, and a 1M context.
84/100
$0.61/1M · 84 community
20
Grok 4.3xAI · 1M ctx · $$$
xAI's flagship — frontier quality at a surprisingly low price, with a 1M context.
84/100
$1.63/1M · 84 community
21
DeepSeek V4 FlashDeepSeek · 1.0M ctx · $
One of the cheapest capable open models — strong for high-volume agent loops.
83/100
$0.18/1M · 83 community
22
Devstral 2Mistral · 262K ctx · $$
Mistral's coding-and-agents specialist — built specifically for SWE-style agent tasks.
83/100
$0.88/1M · 83 community
23
Grok 4.20xAI · 2M ctx · $$$
The 2M-token member of the Grok line — built for very large agent contexts.
83/100
$1.63/1M · 83 community
24
GPT-5.4 MiniOpenAI · 400K ctx · $$$
OpenAI's cheap, fast model for high-volume agent calls.
82/100
$1.88/1M · 82 community
25
GLM 4.7Z.AI · 205K ctx · $$
The refinement of GLM 4.6 — slightly stronger coding at the same friendly price.
82/100
$0.81/1M · 82 community
26
Qwen3 235B A22BAlibaba (Qwen) · 262K ctx · $$
A large open-weight MoE you can self-host — capable and extremely cheap to run hosted.
82/100
$0.23/1M · 82 community
27
Mistral Large 3Mistral · 262K ctx · $$
Mistral's open-weight flagship — a capable European generalist at a fair price.
80/100
$0.80/1M · 80 community
28
MiniMax M2.1MiniMax · 205K ctx · $$
The proven agent workhorse of the MiniMax line — excellent value for tool-heavy runs.
80/100
$0.57/1M · 80 community
29
GLM 4.6Z.AI · 205K ctx · $$
The breakout open-weight coding model — exceptional value that won over the agent community.
80/100
$0.95/1M · 80 community
30
DeepSeek V3.2DeepSeek · 164K ctx · $$
The proven open-weight value model — a staple of budget agent stacks.
77/100
$0.31/1M · 77 community
31
Gemini 3.1 Flash LiteGoogle · 1.0M ctx · $$
One of the cheapest 1M-context models — built for cheap, high-volume calls.
77/100
$0.63/1M · 77 community
32
Llama 4 MaverickMeta · 1.0M ctx · $$
Meta's open-weight MoE — cheap, fast, and a 1M context, with the broadest tooling support.
76/100
$0.38/1M · 76 community
33
DeepSeek R1 (0528)DeepSeek · 164K ctx · $$
The open-weight reasoning model that put long chain-of-thought in everyone's hands.
75/100
$0.99/1M · 75 community
34
CodestralMistral · 256K ctx · $$
Mistral's fast code-completion model — cheap and quick for inline assistance.
72/100
$0.48/1M · 72 community

The model is half the story — the agent is the other half

The model picks the moves; the agent runs the loop, the tools, and the guardrails. Once you've chosen a model, see which agent gets the most out of it.

Compare AI agents →

Every model on this page is included with Standard Compute

Standard Compute

Current frontier models Claude Fable 5 · GPT-5.6 Sol

Your agents never stop.
Your bill never grows.

Frontier models when it counts. Efficient models when it doesn’t. No usage limits. One flat bill.

Free trial · no card · plans from $39/mo

Free trial No credit card needed Billing by Stripe4,900+ agent users

Frequently Asked Questions

For raw quality, Claude Opus 4.8 and GPT-5.5 lead; for the best balance of quality and price, Claude Sonnet 4.6 and GPT-5.4 are the common defaults. For value, open-weight models like DeepSeek V4 Pro, GLM 5.2, and Qwen3.7 win — they deliver near-frontier output at a fraction of the cost. The right pick depends on the agent and the task; our live community votes show what people actually run.