OpenCode Go Models — Chinese AI Model Tracker

Models

Makers

Max Context

9/14

Open Weights

New since April: Kimi K2.6 (Apr 20) · MiMo-V2.5 & V2.5-Pro replace the V2 line (Apr 22) · DeepSeek V4 Pro & Flash (Apr 24) · Qwen3.7 Max, Qwen3.7 Plus & Qwen3.6 Plus join (May 20) · MiniMax M3 (Jun 1). Six-week window, four new ~1T-param sparse-attention flagships with 1M context.

The Go Subscription

Price re-checked Oct 16, 2026 on opencode.ai/go: $10/mo, Go Plus $40/mo. The $5 first-month offer and the limits chart below are from June 9, 2026 and were not re-checked.

$5 first month → $10/mo after

All 14 models, one subscription. Works with OpenCode or any coding agent. Top-up credit available. Cancel any time. Higher tiers (5x → 100x) multiply the limits. Free tier: 200 requests / 5h on Big Pickle + free models.

Subscribe to Go →

Requests per 5 hours — Go 1x tier (log scale)

DeepSeek V4 Flash

31,650

MiMo-V2.5

30,100

Qwen3.7 Plus

4,300

DeepSeek V4 Pro

3,450

MiniMax M2.7

3,400

Qwen3.6 Plus

3,300

MiMo-V2.5-Pro

3,250

MiniMax M3

3,200

Kimi K2.6

1,150

Qwen3.7 Max

950

GLM-5.1

880

Big Pickle + free

200

GLM-5, Kimi K2.5 and MiniMax M2.5 are included but not listed on the published limits chart. The throttle is an inverse capability signal — flagships cost more to serve.

MiniMax

MiniMax M3

New · Jun 1 Multimodal

Newest model in the lineup. MiniMax Sparse Attention (MSA) makes 1M-token context practical — 1/20th per-token compute vs prior gen, 9× faster prefill, 15× faster decode. Native text+image+video input. Open weights promised within days of launch.

ParamsNot disclosed

Context1M (512K min)

ReleasedJun 1, 2026

LicenseOpen (pending)

Key Benchmarks

SWE-Pro 59.0% Terminal-Bench 2.1 66.0% OSWorld 70.0% Beats GPT-5.5 / Gemini 3.1 Pro

$0.60/M in · $2.40/M out

Go: 3,200 req/5h

Alibaba (Qwen)

Qwen3.7 Max

New · May 20 Proprietary

Best SWE-Bench Pro score in the Go lineup (60.6%). Flagship agent model — ran a 35-hour autonomous kernel-optimization session: 1,158 tool calls, 432 evaluations, 10x speedup, zero human intervention. Highest-ranked Chinese model on the AA Intelligence Index (56.6).

ParamsNot disclosed

Context1M / 65K out

ReleasedMay 20, 2026

LicenseProprietary

Key Benchmarks

SWE-Pro 60.6% ★ Terminal-Bench 69.7 MCP-Atlas 76.4 GPQA-D 92.4

$2.50/M in · $7.50/M out

Go: 950 req/5h

Moonshot AI

Kimi K2.6

New · Apr 20 Mod. MIT

Native multimodal agentic flagship. 1T MoE with ~32B active; MLA keeps long-context VRAM cheap. Agent-swarm capable: scales to 300 sub-agents executing 4,000 coordinated steps. Beats Opus 4.6 and GPT-5.4 on SWE-Bench Pro.

Params1T / 32B active

Context256K

ReleasedApr 20, 2026

LicenseModified MIT

Key Benchmarks

SWE-Pro 58.6% HLE w/ tools 54.0% 300-agent swarm

$0.95/M in · $4.00/M out

Go: 1,150 req/5h

DeepSeek

DeepSeek V4 Pro

New · Apr 24 Open Weights

Biggest model in the lineup: 1.6T params, 49B active. CSA+HCA hybrid attention — at 1M context it needs only 27% of the FLOPs and 10% of the KV cache of V3.2. Pro-Max variant posts the highest coding scores of any model (80.6% SWE-Verified, 93.5 LiveCodeBench).

Params1.6T / 49B active

Context1M

ReleasedApr 24, 2026

LicenseOpen Weights

Key Benchmarks

SWE-Verified 80.6% (Max) LiveCodeBench 93.5 MRCR-1M 83.5%

$1.74/M in · $3.48/M out

Go: 3,450 req/5h

DeepSeek

DeepSeek V4 Flash

New · Apr 24 Open Weights

The throughput king of Go — 31,650 requests per 5 hours, the highest allowance in the plan. Efficiency-optimized MoE (284B total, just 13B active) with the same 1M-token context as Pro. The default pick for high-volume agent loops.

Params284B / 13B active

Context1M

ReleasedApr 24, 2026

LicenseOpen Weights

Why It Matters

Highest Go limit 13B active = cheap serving 1M ctx at Flash price

Go limit: 31,650 req/5h

Apr 24, 2026

Xiaomi

MiMo-V2.5-Pro

New · Apr 22 Open Source

Replaces both MiMo-V2-Pro and V2-Omni: one open-source model for reasoning AND full multimodality (image, audio, video, text). 1.02T MoE, hybrid attention + 3-layer multi-token prediction. Uses 40–60% fewer tokens than Opus 4.6 for comparable agentic scores. UltraSpeed variant hits 1,000 tok/s.

Params1.02T / 42B active

Context1M

ReleasedApr 22, 2026

LicenseOpen Source

Key Benchmarks

SWE-Pro 57.2% Token-efficient ClawEval UltraSpeed 1000 tok/s

$1.00/M in · $3.00/M out

Go: 3,250 req/5h

Xiaomi

MiMo-V2.5

New · Apr 22 Multimodal

The efficient sibling in Xiaomi's V2.5 public-beta series (which also includes TTS and ASR variants). Second-highest allowance on Go at 30,100 requests per 5 hours — built to be hammered in agent loops where Pro-class reasoning isn't needed.

ParamsNot disclosed

Context—

ReleasedApr 22, 2026

InputMultimodal

Why It Matters

30,100 req/5h on Go Volume workhorse

Go limit: 30,100 req/5h

Apr 22, 2026

Alibaba (Qwen)

Qwen3.7 Plus

New · May 2026 Vision

Multimodal variant of the 3.7 series with vision input at a lower price point. The budget pick for high-volume, routine workloads — Max is the heavy hitter, Plus is what you run all day. Generous 4,300 req/5h on Go.

ParamsNot disclosed

InputText + Vision

ReleasedMay 2026

LicenseProprietary

Why It Matters

GUI-agent capable 4,300 req/5h on Go

Go limit: 4,300 req/5h

May 2026

Alibaba (Qwen)

Qwen3.6 Plus

Proprietary

Prior-generation Plus, kept in the lineup as a proven high-volume option (3,300 req/5h). Solid general coding and chat at minimal cost — the safe fallback when 3.7 capacity is constrained.

ParamsNot disclosed

Context—

GenerationQwen 3.6

LicenseProprietary

Why It Matters

3,300 req/5h on Go Fallback capacity

Go limit: 3,300 req/5h

2026

Zhipu AI (Z.ai)

GLM-5.1

Reasoning MIT

Long-horizon agentic flagship. Can work autonomously for up to 8 hours on a single task. +20.4pts CyberGym, +7.3pts Terminal-Bench vs GLM-5. Trained entirely on Huawei Ascend chips.

Params744B / ~44B active

Context200K

ReleasedApr 7, 2026

LicenseMIT

Key Benchmarks

SWE-Pro 58.4% CyberGym 68.7% Terminal-Bench 63.5%

API: Z.ai / BigModel.cn

Go: 880 req/5h

Zhipu AI (Z.ai)

GLM-5

Reasoning MIT

First open-weights model to break AA Intelligence Index 50. DeepSeek Sparse Attention for 200K context. Superseded by 5.1 for agentic work but still a strong general reasoner.

Params744B / 44B active

Context200K

ReleasedFeb 11, 2026

LicenseMIT

Key Benchmarks

AA Index 50 SWE-Verified 77.8% AIME 95.4%

API: Contact Z.ai

Feb 11, 2026

Moonshot AI

Kimi K2.5

Multimodal Open Weights

Native multimodal agentic model built via continual pretraining on ~15T visual+text tokens atop K2. MoonViT 400M vision encoder. State-of-the-art visual coding and BrowseComp w/ ctx (74.9%).

ParamsMoE (K2 base)

Context262K

ReleasedJan 27, 2026

LicenseOpen Weights

Key Benchmarks

HLE w/ tools 50.2% AIME 96.1% BrowseComp 74.9%

$0.38/M in · $1.72/M out

Jan 27, 2026

MiniMax

MiniMax M2.7

Reasoning Not Yet Open

First model deeply participating in its own evolution. Self-improving with Agent Teams and dynamic tool search. Highest GDPval-AA Elo (1495) among open-source models. 97% skill adherence.

ParamsNot disclosed

Context200K

ReleasedMar 18, 2026

LicenseTBD

Key Benchmarks

AA Index 50 GDPval-AA 1495 Elo SWE-Pro 56.2%

$0.30/M in · $1.20/M out

Go: 3,400 req/5h

MiniMax

MiniMax M2.5

Reasoning MIT

Open-source workhorse with best-in-class SWE-Bench Verified (80.2%) at launch. ~100 tok/s inference. Matching Opus-class coding at 1/20th the cost — still the value benchmark of the lineup.

Params~229B / 10B active

Context200K

ReleasedFeb 12, 2026

LicenseMIT

Key Benchmarks

SWE-Verified 80.2% Multi-SWE 51.3% BrowseComp 76.3%

$0.15/M in · $1.20/M out

Feb 12, 2026

Quick Comparison

Model	Maker	Params	Context	License	Best At	Price (in/out)	Go req/5h
Qwen3.7 Max	Alibaba	—	1M	Proprietary	SWE-Pro (60.6%), 35h autonomy	$2.50/$7.50	950
MiniMax M3	MiniMax	—	1M	Open (pending)	SWE-Pro 59.0%, MSA speed	$0.60/$2.40	3,200
Kimi K2.6	Moonshot	1T/32B	256K	Mod. MIT	Agent swarm, multimodal coding	$0.95/$4.00	1,150
DeepSeek V4 Pro	DeepSeek	1.6T/49B	1M	Open Wt.	Coding (80.6% SWE-V, Max)	$1.74/$3.48	3,450
DeepSeek V4 Flash	DeepSeek	284B/13B	1M	Open Wt.	Throughput, volume agents	—	31,650
MiMo-V2.5-Pro	Xiaomi	1.02T/42B	1M	Open Source	Token efficiency, multimodal	$1/$3	3,250
MiMo-V2.5	Xiaomi	—	—	—	Volume workhorse	—	30,100
Qwen3.7 Plus	Alibaba	—	—	Proprietary	Vision, budget volume	—	4,300
Qwen3.6 Plus	Alibaba	—	—	Proprietary	Fallback capacity	—	3,300
GLM-5.1	Zhipu AI	~744B/44B	200K	MIT	Long-horizon (8h autonomy)	Z.ai API	880
GLM-5	Zhipu AI	744B/44B	200K	MIT	AA Index, SWE-Verified	Contact Z.ai	—
Kimi K2.5	Moonshot	MoE	262K	Open Wt.	Multimodal, BrowseComp	$0.38/$1.72	—
MiniMax M2.7	MiniMax	—	200K	TBD	Self-evolution, GDPval-AA	$0.30/$1.20	3,400
MiniMax M2.5	MiniMax	229B/10B	200K	MIT	SWE-Verified (80.2%), Cost	$0.15/$1.20	—

Go req/5h = requests per 5-hour window on the Go 1x tier ($10/mo). Higher tiers (5x–100x) multiply these limits. "—" = included in Go but not on the published limits chart. API prices are list prices on OpenRouter/Fireworks/first-party, not what Go charges.

Command Go Command Code · Go plan · checked Oct 16, 2026

Assumption: "Command Go" = the Go plan of Command Code (commandcode.ai), the $1/month entry plan of the command-code CLI. If you meant a different service, this section is the only part to change. All figures from commandcode.ai/docs/plans/go and commandcode.ai/pricing, checked Oct 16, 2026.

Item	OpenCode Go	Command Go
Price	$10/mo (Go Plus $40/mo)	$1/mo, buys $10 of credits
Usage limits	Per-model monthly usage + est. requests/5h	$3 / 5h · $6 / week · $10 / month (shared pool; per-model credits $6–$10 since Sep 28, 2026)
API access	Use with any agent	No (Go is the only Command Code plan without API access; GOAT $10/mo adds it)
Client	OpenCode or any agent	`npm i -g command-code`, run `cmd`
Free models	LongCat 2.5 Preview, Beam (limited time)	Laguna S 2.1, Glyph Cluster, Cat Jam

Command Go: models with boosted credits (vendor list prices per 1M tokens)

Model	Input	Output	Monthly credits on Go	Est. req / 5h
DeepSeek V4.1 Flash	$0.15	$0.60	$10	7,690
GLM-5.2	$1.40	$4.40	$10	203
GLM-5.3	$1.40	$4.40	$10	203
GLM-5.3 Flash	$0.15	$0.50	$10	1,770
Tencent Hy3	$0.14	$0.58	$10	1,520
Qwen 3.8 27B	$0.40	$3.00	$10	1,030
Qwen 3.7 Max	$2.50	$7.50	$10	105
Qwen 3.7 Plus	$0.40	$1.60	$10	647
Qwen 3.6 Plus	$0.50	$3.00	$10	500
Mistral Large 4	$1.36	$4.18	$6 ($10 through Oct 20)	336
LongCat 2.0	$0.30	$1.20	$10	3,850
Kimi K2.7 Code	$0.95	$4.00	$9	244
MiniMax M3	$0.30	$1.20	$8	708
Kimi K3	$3.00	$15.00	$8	118
Qwen 3.8 Max	$2.00	$6.00	$7	137
Grok 4.5	$2.00	$6.00	$6	64.7

About 40 more models (DeepSeek V4 Pro/Flash, MiMo V2.5/V2.6, GPT-6 Luna, Muse Spark, Step, Nemotron…) carry the standard $6 credits. Request estimates are Command Code's own, for a typical agent request (~800 new input, ~50K cached, 125–200 output tokens). Source: commandcode.ai/docs/plans/go, checked Oct 16, 2026.

New: Mistral & Grok (via Grok Bot)

Mistral models, free tiers and API prices, Grok models and xAI API pricing, and when to pick each vs OpenCode Go and Command Go. freeintelligence.ai/mistral-grok →

Get OpenCode Go — $10/mo →

Data sourced from opencode.ai/go, official releases, HuggingFace, OpenRouter, and Artificial Analysis. Model cards and Quick Comparison last updated June 9, 2026. OpenCode Go price ($10/mo, Go Plus $40/mo) re-checked on opencode.ai/go and Command Go section added Oct 16, 2026. See also Mistral & Grok.

Built with here.now