Standard Compute

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. 2 community ratings so far. Add yours.

Updated October 13, 2026. New in the rankings: Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, GPT-6.1 Sol, GPT-6 Astra, GPT-6 Luna, Gemini 3.8 Flash, GLM 5.3, DeepSeek V4.1 Flash, Grok 4.7. Community ratings take time to build after a release.

Leaderboards: Smartest · Best for coding · Fastest · Lowest latency · Best value · Cheapest per request

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 per dollar to rank by quality per dollar instead. Tap any model for the full breakdown.

Rank by
Show
1
Claude Opus 5.5Anthropic · 1M ctx
Anthropic's flagship since September 22, 2026, and #1 on the tbench.ai Terminal-Bench 4.0 leaderboard at 64.9% in Claude Code.
98/100
$8.80/1M
2
Claude Sonnet 5.5Anthropic · 1M ctx
Anthropic's mid-tier model since September 28, 2026: 61.8% on the tbench.ai Terminal-Bench 4.0 leaderboard at half the price of Opus 5.5.
96/100
$4.40/1M
3
Claude Fable 5.1Anthropic · 1M ctx
Anthropic's premium tier since September 1, 2026, at $10/$50 per million tokens; Anthropic says Opus 5.5 now matches it on most work.
93/100
$22/1M
4
GPT-6 AstraOpenAI · 1.1M ctx
OpenAI's top GPT-6 tier since September 3, 2026, at $10/$50 per million tokens.
92/100
$22/1M
5
Gemini 4 ArgonGoogle · 1.0M ctx
Google's first Gemini 4 model, a flagship reasoning model at an introductory $2/$10 per million tokens that doubles on December 1, 2026.
92/100
$4.40/1M
6
GPT-6.1 SolOpenAI · 1.1M ctx
OpenAI's recommended everyday model since September 29, 2026: it ties Astra on the tbench.ai Terminal-Bench 4.0 leaderboard at a fifth of the price.
92/100
$4.40/1M
7
Muse WonderMeta · 1.0M ctx
Meta's frontier reasoning model since October 12, 2026, scoring 52 on the Artificial Analysis Intelligence Index at $4/$20 per million tokens.
90/100
$8.80/1M
8
GPT-6 SolOpenAI · 1.1M ctx
OpenAI's mid tier from September 22, 2026, replaced a week later by GPT-6.1 Sol at the same price.
88/100
$4.40/1M
9
Claude Fable 5Anthropic · 1M ctx
Anthropic's June 2026 premium model, superseded by Claude Fable 5.1 (September 2026) and still on the API.
87/100
$22/1M · 98 community (1 rating)
10
GPT-5.6 SolOpenAI · 1.1M ctx
OpenAI's July 2026 top tier, superseded by GPT-6.1 Sol (September 2026) and still on the API.
84/100
$4.40/1M
11
GLM 5.3Z.AI · 1.0M ctx
Z.AI's open-weight flagship (API August 14, weights August 29, 2026), with the top open-weight score on the tbench.ai Terminal-Bench 4.0 leaderboard (41.8%).
83/100
$1.50/1M
12
Grok 4.7xAI · 500K ctx
xAI's current model since September 21, 2026, with a 500K context and $6 per million output tokens.
82/100
$3.20/1M
13
Claude Haiku 5.5Anthropic · 1M ctx
Anthropic's small model since October 7, 2026, at $0.10/$0.50 per million tokens, with a 5x price step above 100K tokens.
82/100
$0.22/1M
14
GPT-5.6 TerraOpenAI · 1.1M ctx
The middle tier of OpenAI's July 2026 GPT-5.6 family, at $2/$12 per million tokens.
82/100
$5.00/1M
15
Step 5 PreviewStepFun · 1M ctx
StepFun's agent-focused flagship, on OpenRouter since October 8, 2026, scoring 44 on the Artificial Analysis Intelligence Index at $1/$2.70 per million tokens.
80/100
$1.51/1M
16
DeepSeek V4.1 FlashDeepSeek · 1.0M ctx
DeepSeek's MIT-licensed model since September 10, 2026, and #1 by tokens on OpenCode's public usage board in October 2026.
77/100
$0.22/1M
17
DeepSeek V4 FlashDeepSeek · 1.0M ctx
DeepSeek's original V4 Flash; DeepSeek's API now serves this name with V4.1 Flash.
77/100
$0.12/1M
18
Gemini 3.8 FlashGoogle · 1.0M ctx
Google's fast everyday model since September 2, 2026, at an introductory price that doubles in January.
76/100
$1.65/1M
19
Mistral Large 4Mistral · 1.0M ctx
Mistral's open-weight flagship since October 6, 2026: 1M context and image input, sold at half its $1.36/$4.18 list price for now.
76/100
$1.10/1M
20
Claude Opus 4.8Anthropic · 1M ctx
Anthropic's previous Opus, superseded by Claude Opus 5.5 (September 2026) and still available on the API.
76/100
$11/1M
21
GPT-6 LunaOpenAI · 1.1M ctx
OpenAI's small, fast GPT-6 model since September 22, 2026, for focused, high-volume tasks.
73/100
$0.22/1M
22
GPT-5.5OpenAI · 1.1M ctx
OpenAI's previous flagship; retires from ChatGPT and Codex sign-in on October 14, 2026, stays on the API.
73/100
$13/1M
23
GPT-5.4OpenAI · 1.1M ctx
OpenAI's March 2026 GPT-5.4, at $2.50/$15 per million tokens, superseded by the GPT-6 family.
73/100
$6.25/1M
24
GPT-5.6 LunaOpenAI · 1.1M ctx
The fast, cheap tier of the GPT-5.6 family, superseded by GPT-6 Luna (September 2026).
71/100
$0.50/1M
25
DeepSeek V4 ProDeepSeek · 1.0M ctx
DeepSeek's larger open-weight V4 model, updated August 2026, with a 1M-token context.
71/100
$0.56/1M
26
Gemini 3.5 FlashGoogle · 1.0M ctx
Google's previous Flash default for high-throughput agents, superseded by Gemini 3.8 Flash (September 2026).
67/100
$3.75/1M
27
GPT-5.3-CodexOpenAI · 400K ctx
OpenAI's coding-tuned model from February 2026, now deprecated in Codex with ChatGPT sign-in.
65/100
$5.42/1M
28
Gemini 3.1 ProGoogle · 1.0M ctx
Google's Gemini 3.1 Pro, still a preview model, with a 1M-token multimodal context.
64/100
$5.00/1M
29
Muse Spark OpenMeta · 1.0M ctx
Meta's open-weight model for coding and agent work, released October 15, 2026: 816B parameters with 42B active, a 1M-token context and the Muse Model License.
64/100
$2.15/1M
30
Beam 1.0Reflection · 1.0M ctx
Reflection's open-weight model for coding and agent work, released October 16, 2026: 501B parameters with 23B active, a 1M-token context and an Apache 2.0 license.
64/100
$1.60/1M
31
GLM 5.2Z.AI · 1.0M ctx
Z.AI's previous open-weight flagship, superseded by GLM 5.3 (August 2026).
63/100
$1.72/1M · 86 community (1 rating)
32
Qwen3.7 MaxAlibaba (Qwen) · 1M ctx
Alibaba's hosted Qwen3.7 Max from May 2026, with a 1M-token context.
63/100
$2.36/1M
33
MiniMax M3MiniMax · 1.0M ctx
MiniMax's June 2026 open-weight model for agents, with a 1M context.
63/100
$0.57/1M
34
Claude Sonnet 4.6Anthropic · 1M ctx
Anthropic's February 2026 Sonnet, superseded by Claude Sonnet 5.5 (September 2026).
62/100
$6.60/1M
35
GPT-5.4 MiniOpenAI · 400K ctx
OpenAI's small GPT-5.4 model from March 2026, at $0.75/$4.50 per million tokens.
61/100
$1.88/1M
36
Grok 4.20xAI · 2M ctx
The 2M-token member of the Grok line, built for very large contexts.
59/100
$1.63/1M
37
Kimi K2.7 CodeMoonshot AI · 262K ctx
Moonshot AI's coding-tuned open-weight model from June 2026, with a 262K context.
59/100
$1.48/1M
38
Kimi K2.6Moonshot AI · 262K ctx
Moonshot AI's April 2026 open-weight general model, with a 262K context.
59/100
$1.48/1M
39
Qwen3.7 PlusAlibaba (Qwen) · 1M ctx
Alibaba's lower-cost hosted Qwen3.7 model, with a 1M-token context.
58/100
$0.61/1M
40
Grok 4.3xAI · 1M ctx
xAI's April 2026 model with a 1M context, superseded by Grok 4.7 (September 2026).
58/100
$1.63/1M
41
DeepSeek V3.2DeepSeek · 164K ctx
DeepSeek's December 2025 open-weight model, superseded by the V4 line.
57/100
$0.42/1M
42
GLM 4.7Z.AI · 205K ctx
Z.AI's December 2025 update to GLM 4.6, superseded by the GLM 5 line.
57/100
$1.08/1M
43
MiniMax M2.1MiniMax · 205K ctx
MiniMax's December 2025 open-weight agent model, superseded by MiniMax M3.
56/100
$0.57/1M
44
Gemini 3.1 Flash LiteGoogle · 1.0M ctx
Google's budget Flash-Lite model from March 2026, built for simple, high-volume calls.
55/100
$0.63/1M
45
GLM 4.6Z.AI · 205K ctx
Z.AI's September 2025 open-weight coding model, superseded by GLM 4.7 and the GLM 5 line.
54/100
$0.95/1M
46
Claude Haiku 4.5Anthropic · 200K ctx
Anthropic's previous small model, superseded by Claude Haiku 5.5 (October 2026).
53/100
$2.20/1M
47
CodestralMistral · 256K ctx
Mistral's code-completion model, built for autocomplete and quick edits.
53/100
$0.48/1M
48
DeepSeek R1 (0528)DeepSeek · 164K ctx
DeepSeek's May 2025 open-weight reasoning model, now well behind current models.
50/100
$0.99/1M
49
Qwen3 235B A22BAlibaba (Qwen) · 262K ctx
Alibaba's July 2025 open-weight mixture-of-experts model, with 22B active parameters.
50/100
$0.23/1M
50
Devstral 2Mistral · 262K ctx
Mistral's open-weight model for software-engineering agents, from December 2025.
50/100
$0.88/1M
51
Mistral Large 3Mistral · 262K ctx
Mistral's open-weight generalist from December 2025, with a 262K context and EU provenance.
49/100
$0.80/1M
52
Llama 4 MaverickMeta · 1.0M ctx
Meta's April 2025 open-weight mixture-of-experts model, with a 1M context and wide tooling support.
49/100
$0.33/1M

The agent matters as much as the model. Compare AI agents →

Explore the models available through Standard Compute

Standard Compute

Current frontier models Claude Fable 5.1 · GPT-6 Astra

More intelligence
per dollar.

Each step goes to the cheapest model that can do it, on providers we vet for price, uptime and speed, with automatic fallback when one fails.

Compliance
Opt-in rules keep a company, team or key on European or US providers
Privacy
We don’t train on your prompts or store them

Fixed monthly plans from $19 · cancel anytime

7-day fair refund Free trial, no card Billing by Stripe4,100+ agent users

Frequently Asked Questions

As of October 2026, Claude Opus 5.5 leads the tbench.ai Terminal-Bench 4.0 leaderboard (64.9% in Claude Code) and has the highest Artificial Analysis Intelligence Index score of the models we track, with Claude Sonnet 5.5 close behind at half the price. GPT-6.1 Sol ties GPT-6 Astra on the tbench.ai leaderboard (58.2%) at a fifth of Astra's price. Two October releases join the frontier group: Google's Gemini 4 Argon (57.1% in Artificial Analysis's Terminal-Bench 4.0 run) and Meta's Muse Wonder (52.0%). For cheap, high-volume steps, Claude Haiku 5.5 and GPT-6 Luna both list at $0.10/$0.50 per million tokens. GLM 5.3 and DeepSeek V4.1 Flash are the leading open-weight options. Compare models on your own tasks and budget; newly released models may not yet have community ratings.