devadmin/OpenCodeGoEvals
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

chore: weekly evaluation refresh 2026-11-02

Browse Source
This commit is contained in:
opencode-go-evals
2026-11-02 05:24:14 +00:00
parent f3259fdc0a
commit 316141d80d
6 changed files with 1962 additions and 1702 deletions
Download Patch File Download Diff File Expand all files Collapse all files
MODELS.md
+306 -306
View File
@@ -5,7 +5,7 @@ Short, practical notes on every model in the OpenCode Go catalogue: what each on
what it costs, and what to watch out for.
Evaluated scores come from Artificial Analysis (single consistent harness) plus vendor model cards.
Catalogue metadata comes from models.dev's OpenCode Go provider. Data snapshot: **2026-10-26**.
Catalogue metadata comes from models.dev's OpenCode Go provider. Data snapshot: **2026-11-02**.
The interactive version with the full task matrix lives at `site/index.html`.
> No model in the Go catalogue generates images. Every Go model is text-output; several accept image,
@@ -28,13 +28,13 @@ The interactive version with the full task matrix lives at `site/index.html`.
### undefined
**moonshotai** · released 2026-10-23 · $2 in / $10 out per 1M tokens · 1.05M context · text, image, video in → text out
**google** · released 2026-10-13 · $2 in / $10 out per 1M tokens · 1.05M context · text, image, video, audio, pdf in → text out
Multimodal Kimi model with 1M context and always-on reasoning for coding, 3D creation, financial research, and agentic work
Google's flagship reasoning model for complex software engineering, deep math and science, and long-horizon autonomous agents
_No hand-written write-up yet._
*Evaluated:* Intelligence Index 55.1 · FrontierCode 47.1 · TB4.0 67.2 · AutomationBench 73.3 · HLE 52.5 · LCR 86.3 · MMMU-Pro 83.9
*Evaluated:* Intelligence Index 50.4 · FrontierCode 43.5 · TB4.0 57.1 · AutomationBench 77.5 · HLE 57.1 · LCR 79.7
### undefined
@@ -46,7 +46,7 @@ Google's flagship reasoning model for complex software engineering, deep math an
Hardest frontier work: large codebase refactors and optimisation (Google used it to optimise a 200,000-line Python codebase overnight), financial research, legal drafting, deep math/science, and autonomous security work such as vulnerability patching. It is also the model to reach for when a task needs to stay coherent across a very long autonomous run — the 1M window and long max-output budgets are the point, not the per-token price.
*Evaluated:* Intelligence Index 52.6 · FrontierCode 43.5 · TB4.0 57.1 · AutomationBench 77.5 · HLE 57.1 · LCR 79.7
*Evaluated:* Intelligence Index 50.4 · FrontierCode 43.5 · TB4.0 57.1 · AutomationBench 77.5 · HLE 57.1 · LCR 79.7
**Caveats**
@@ -56,13 +56,13 @@ Hardest frontier work: large codebase refactors and optimisation (Google used it
### undefined
**google** · released 2026-10-13 · $2 in / $10 out per 1M tokens · 1.05M context · text, image, video, audio, pdf in → text out
**moonshotai** · released 2026-10-23 · $2 in / $10 out per 1M tokens · 1.05M context · text, image, video in → text out
Google's flagship reasoning model for complex software engineering, deep math and science, and long-horizon autonomous agents
Multimodal Kimi model with 1M context and always-on reasoning for coding, 3D creation, financial research, and agentic work
_No hand-written write-up yet._
*Evaluated:* Intelligence Index 52.6 · FrontierCode 43.5 · TB4.0 57.1 · AutomationBench 77.5 · HLE 57.1 · LCR 79.7
*Evaluated:* Intelligence Index 46 · FrontierCode 47.1 · TB4.0 67.2 · AutomationBench 73.3 · HLE 52.5 · LCR 86.3 · MMMU-Pro 83.9
### undefined
@@ -74,7 +74,7 @@ Meta's multimodal reasoning model for long-running agentic, multi-agent and codi
Multi-agent and long-horizon pipelines: several specialised agents handing work to each other over a 1M window, with the widest input surface in Go (image, video, PDF and audio all accepted). Cheap enough ($0.10 / $0.20) to keep several agents running at once — it is the pick when the architecture, not the single call, is doing the work.
*Evaluated:* Intelligence Index 48.1 · FrontierCode 40.6 · TB2.1 84.3 · AutomationBench 57.9 · GPQA 93.5 · HLE 48.7 · LCR 83
*Evaluated:* Intelligence Index 36.4 · FrontierCode 40.6 · TB2.1 84.3 · AutomationBench 57.9 · GPQA 93.5 · HLE 48.7 · LCR 83
**Caveats**
@@ -83,6 +83,23 @@ Multi-agent and long-horizon pipelines: several specialised agents handing work
### undefined
**anthropic** · released 2026-10-07 · $0.1 in / $0.5 out per 1M tokens · 1M context · text, image, pdf in → text out
Targeted for: **Fast agents, Classification, Responsive UX, Cheap**
Fast Claude model for responsive assistance, classification, and lightweight agents.
Anything that has to answer fast and often: interactive completions, classification and extraction, lightweight sub-agents, and UI-style assistance where latency matters more than depth. Anthropic's tool-calling behaviour is the draw for lightweight agents.
*Evaluated:* Intelligence Index 36.2 · FrontierCode 41.8 · TB4.0 32.8 · AutomationBench 35.4 · HLE 44.4 · LCR 82.7
**Caveats**
- Price steps up above 100K tokens of context (5×).
- Requests are retained for 30 days under Anthropic's data policy — not zero-data-retention.
### undefined
**xai** · released 2026-09-21 · $2 in / $6 out per 1M tokens · 0.50M context · text, image, pdf in → text out
Targeted for: **Long-running agents, Coding, Knowledge work, Visual projects**
@@ -91,7 +108,7 @@ xAI's frontier model for long-running agents, coding, knowledge work, and visual
Sustained agent sessions that mix code and document/visual inputs — research that ends in a written deliverable, codebase work with images in the loop, or an analysis that has to survive a long run without losing the thread. Priced at the top of the open-model tier but with a 500K window that suits multi-file, multi-hour tasks.
*Evaluated:* Intelligence Index 46.4 · FrontierCode 33.8 · TB4.0 25.8 · AutomationBench 65.6 · HLE 43.1 · LCR 76.7
*Evaluated:* Intelligence Index 34.3 · FrontierCode 33.8 · TB4.0 25.8 · AutomationBench 65.6 · HLE 43.1 · LCR 76.7
**Caveats**
@@ -100,23 +117,6 @@ Sustained agent sessions that mix code and document/visual inputs — research t
### undefined
**xiaomi** · released 2026-09-22 · $0.435 in / $0.87 out per 1M tokens · 1.05M context · text, image, audio, video in → text out
Targeted for: **Coding agents, Multimodal, Automation, Long context**
Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution.
Coding agents that also have to look at things: screenshots, video, audio alongside code. Xiaomi's own card highlights strong AutomationBench and terminal scores, so it is aimed at long mechanical agent loops as much as at authoring code.
*Evaluated:* Intelligence Index 46.3 · FrontierCode 45.4 · TB4.0 34.8 · AutomationBench 58.6 · HLE 49.4 · LCR 86.3
**Caveats**
- Text output only.
- The Free tier of this family (MiMo-V2.6-Flash) sits outside the Go plan — check the default provider list.
### undefined
**alibaba** · released 2026-08-03 · $2 in / $6 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Coding, Professional work, Multimodal, Long-horizon agents**
@@ -125,7 +125,7 @@ A 2.4-trillion-parameter MoE flagship for coding, professional work, multimodal
Broad, heavy-duty work where a single model has to cover the whole job: writing code, producing professional deliverables (the GDPval/JobBench scores are strong), and reasoning over images, video and PDFs in the same session. Also the strongest general agent in Go on the terminal and tool-use benchmarks.
*Evaluated:* Intelligence Index 45.4 · FrontierCode 43.2 · TB2.1 88.8 · AutomationBench 56.2 · GPQA 92.8 · HLE 43.1 · LCR 80.3 · MMMU-Pro 82.8
*Evaluated:* Intelligence Index 33.6 · FrontierCode 43.2 · TB2.1 88.8 · AutomationBench 56.2 · GPQA 92.8 · HLE 43.1 · LCR 80.3 · MMMU-Pro 82.8
**Caveats**
@@ -142,7 +142,7 @@ Flagship GLM model for long-horizon coding, agents, and complex project delivery
Owning a project rather than a ticket: sustained multi-file work, terminal and repo tasks, and the security-agent track (CyberGym, ExploitGym, ExploitBench) where GLM scores are among the strongest in Go. Text-only, and one of the pricier models, so it is aimed at the tasks that genuinely need a flagship.
*Evaluated:* Intelligence Index 44.8 · FrontierCode 40.1 · TB2.1 83.9 · AutomationBench 62.2 · GPQA 91.7 · HLE 42.3 · LCR 79.7
*Evaluated:* Intelligence Index 31.7 · FrontierCode 40.1 · TB2.1 83.9 · AutomationBench 62.2 · GPQA 91.7 · HLE 42.3 · LCR 79.7
**Caveats**
@@ -151,65 +151,20 @@ Owning a project rather than a ticket: sustained multi-file work, terminal and r
### undefined
**stepfun** · released 2026-09-16 · $1 in / $2.7 out per 1M tokens · 1M context · text, image, video in → text out
**xiaomi** · released 2026-09-22 · $0.435 in / $0.87 out per 1M tokens · 1.05M context · text, image, audio, video in → text out
Targeted for: **Coding, Professional knowledge work, Multimodal, Open weights**
Targeted for: **Coding agents, Multimodal, Automation, Long context**
StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image and video input and a 1M-token context window.
Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution.
Coding plus deliverables: a base model strong enough to write production code and also produce the spreadsheets, reports and documents that surround it. Native image and video input support it works from screenshots and clips directly.
Coding agents that also have to look at things: screenshots, video, audio alongside code. Xiaomi's own card highlights strong AutomationBench and terminal scores, so it is aimed at long mechanical agent loops as much as at authoring code.
*Evaluated:* Intelligence Index 43.7 · FrontierCode 38.4 · TB4.0 33.3 · AutomationBench 51 · HLE 46.5 · LCR 88.3 · MMMU-Pro 76.4
*Evaluated:* Intelligence Index 29.9 · FrontierCode 45.4 · TB4.0 34.8 · AutomationBench 58.6 · HLE 49.4 · LCR 86.3
**Caveats**
- Preview release — expect the numbers to move.
- Currently priced with a promotional 4× allowance for a limited time.
### undefined
**moonshotai** · released 2026-07-16 · $3 in / $15 out per 1M tokens · 1.05M context · text, image, video in → text out
Targeted for: **Long-horizon agents, Coding, Web research, Spreadsheets, Open weights**
Multimodal Kimi model with a 1M context and toggleable max-effort thinking for long-horizon agent work.
The Kimi tier built for endurance: long-horizon agent runs, web research (BrowseComp), coding (FrontierSWE, DeepSWE, Terminal-Bench), and office deliverables — it scores well on SpreadsheetBench, which makes it the pick for data-analysis and spreadsheet work in Go. Thinking effort can be turned up for hard problems or down for speed.
*Evaluated:* Intelligence Index 43.6 · FrontierCode 43.6 · TB2.1 85 · AutomationBench 58.3 · GPQA 93.5 · HLE 46.9 · LCR 88.7 · MMMU-Pro 80.5
**Caveats**
- Most expensive model in Go at $3 / $15 per 1M tokens.
- Terminal-Bench 4.0 is much weaker than its Terminal-Bench 2.1 result — hard-shell tasks still cost it a lot.
- Text output only.
### undefined
**anthropic** · released 2026-10-07 · $0.1 in / $0.5 out per 1M tokens · 1M context · text, image, pdf in → text out
Targeted for: **Fast agents, Classification, Responsive UX, Cheap**
Fast Claude model for responsive assistance, classification, and lightweight agents.
Anything that has to answer fast and often: interactive completions, classification and extraction, lightweight sub-agents, and UI-style assistance where latency matters more than depth. Anthropic's tool-calling behaviour is the draw for lightweight agents.
*Evaluated:* Intelligence Index 43.4 · FrontierCode 41.8 · TB4.0 32.8 · AutomationBench 35.4 · HLE 44.4 · LCR 82.7
**Caveats**
- Price steps up above 100K tokens of context (5×).
- Requests are retained for 30 days under Anthropic's data policy — not zero-data-retention.
### undefined
**minimax** · released 2026-10-19 · $0.15 in / $0.6 out per 1M tokens · 1M context · text, image, video in → text out
Frontier multimodal coding model with 1M context window and tunable thinking depth
_No hand-written write-up yet._
*Evaluated:* Intelligence Index 42.7 · FrontierCode 38.6 · TB4.0 25.8 · AutomationBench 72.5 · HLE 48 · LCR 79.7 · MMMU-Pro 79.1
- The Free tier of this family (MiMo-V2.6-Flash) sits outside the Go plan — check the default provider list.
### undefined
@@ -221,7 +176,7 @@ Frontier multimodal coding model with a 1M-token context window and tunable thin
Day-to-day coding-agent work on real repositories: reading long codebases, editing files, running tools. Image and video input make it usable for UI/screenshot-driven work (reading a rendered page, describing an asset). Because thinking depth is tunable, it can be dialed down for cheap, high-volume turns and up for hard problems.
*Evaluated:* Intelligence Index 42.7 · FrontierCode 38.6 · TB4.0 25.8 · AutomationBench 72.5 · HLE 48 · LCR 79.7 · MMMU-Pro 79.1
*Evaluated:* Intelligence Index 28.9 · FrontierCode 38.6 · TB4.0 25.8 · AutomationBench 72.5 · HLE 48 · LCR 79.7 · MMMU-Pro 79.1
**Caveats**
@@ -230,6 +185,33 @@ Day-to-day coding-agent work on real repositories: reading long codebases, editi
### undefined
**minimax** · released 2026-10-19 · $0.15 in / $0.6 out per 1M tokens · 1M context · text, image, video in → text out
Frontier multimodal coding model with 1M context window and tunable thinking depth
_No hand-written write-up yet._
*Evaluated:* Intelligence Index 28.9 · FrontierCode 38.6 · TB4.0 25.8 · AutomationBench 72.5 · HLE 48 · LCR 79.7 · MMMU-Pro 79.1
### undefined
**stepfun** · released 2026-09-16 · $1 in / $2.7 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Coding, Professional knowledge work, Multimodal, Open weights**
StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image and video input and a 1M-token context window.
Coding plus deliverables: a base model strong enough to write production code and also produce the spreadsheets, reports and documents that surround it. Native image and video input support it works from screenshots and clips directly.
*Evaluated:* Intelligence Index 28.5 · FrontierCode 38.4 · TB4.0 33.3 · AutomationBench 51 · HLE 46.5 · LCR 88.3 · MMMU-Pro 76.4
**Caveats**
- Preview release — expect the numbers to move.
- Currently priced with a promotional 4× allowance for a limited time.
### undefined
**zhipuai** · released 2026-08-26 · $0.15 in / $0.5 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Efficient coding, Long-horizon agents, Multimodal, Cheap**
@@ -238,7 +220,7 @@ Native multimodal GLM model for efficient coding and long-horizon agent tasks.
The cheap twin of the flagship: same family behaviour, multimodal input, and one of the better Terminal-Bench scores in Go per dollar. It is the sensible pick for long agent loops that need to see images, documents or video but do not need frontier depth.
*Evaluated:* Intelligence Index 41.8 · FrontierCode 41 · TB2.1 84.3 · AutomationBench 60.4 · GPQA 91.2 · HLE 39.9 · LCR 80
*Evaluated:* Intelligence Index 27.8 · FrontierCode 41 · TB2.1 84.3 · AutomationBench 60.4 · GPQA 91.2 · HLE 39.9 · LCR 80
**Caveats**
@@ -247,57 +229,6 @@ The cheap twin of the flagship: same family behaviour, multimodal input, and one
### undefined
**alibaba** · released 2026-08-26 · $0.15 in / $0.47 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Visual reasoning, Documents, Agent tasks, Cheap**
Qwen vision-language model for visual reasoning, documents, and agent tasks.
Perception-heavy work at a low price: reading screenshots and diagrams, extracting from documents, and driving agents that need to see what is on screen. It pairs a 1M window with sub-20¢ input pricing, which makes it a good bulk worker for visual pipelines.
*Evaluated:* Intelligence Index 39.8 · TB2.1 86.1 · AutomationBench 55.9 · GPQA 92.3 · HLE 38 · LCR 79.7 · MMMU-Pro 79.8
**Caveats**
- Text output only — no image generation.
- Evaluated publicly as 'Qwen3.8-Flash-Next' in third-party benchmark tables.
### undefined
**meta** · released 2026-08-05 · $0.1 in / $0.2 out per 1M tokens · 1.05M context · text, image, video, pdf, audio in → text out
Targeted for: **Coding, Multimodal, Cheap**
Coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.
The previous Muse generation — kept in Go because it is cheap and coding-focused. Pick it when you want Meta's multimodal input surface at $0.10/$0.20 and 1.3 is not required; otherwise use Muse Spark 1.3.
*Evaluated:* Intelligence Index 39.6 · TB2.1 80.1 · AutomationBench 40.6 · GPQA 90.4 · HLE 45.5 · LCR 79
**Caveats**
- Contributor tier — prompts and completions may be used to train future Meta models.
- Superseded by Muse Spark 1.3 on long-horizon agent tasks.
### undefined
**deepseek** · released 2026-09-10 · $0.15 in / $0.6 out per 1M tokens · 1M context · text, image in → text out
Targeted for: **Agentic coding, Reasoning, Cheap, High volume**
DeepSeek's V4.1 Flash: built for reasoning and agentic coding, and the single most-used model in OpenCode by token volume.
The default engine for long coding-agent runs. DeepSeek's own card lists Terminal-Bench, DeepSWE, CyberGym and AutomationBench scores, i.e. it is tuned for terminal work, repo-level fixes and security tasks rather than chat. With cache reads at ~0.3¢/1M it is also the cheapest way to move large volumes of agent traffic.
*Evaluated:* Intelligence Index 39.5 · FrontierCode 44 · TB4.0 26.8 · AutomationBench 68.9 · HLE 39.2 · LCR 84 · MMMU-Pro 77
**Caveats**
- Text output only.
- Price doubles during DeepSeek's peak hours (01:00–04:00 and 06:00–10:00 UTC, Mon–Fri).
### undefined
**openai** · released 2026-09-22 · $0.1 in / $0.5 out per 1M tokens · 1.05M context · text, image, pdf in → text out
Targeted for: **High-volume coding, Agents, Document work, Cheap**
@@ -306,7 +237,7 @@ OpenAI's most efficient model for focused, high-volume tasks, at the price of a
The default workhorse when you want GPT-class behaviour without GPT-class spend: everyday coding-agent turns, small refactors, summarising long documents (1.05M window), classifying and routing. In Go it is one of the best price/performance points, so it is a good default for background agents and a fallback when a heavier model is unnecessary.
*Evaluated:* Intelligence Index 38.1 · FrontierCode 41.9 · TB4.0 12.6 · AutomationBench 53.2 · HLE 38.5 · LCR 83.3 · MMMU-Pro 79.7
*Evaluated:* Intelligence Index 27.8 · FrontierCode 41.9 · TB4.0 12.6 · AutomationBench 53.2 · HLE 38.5 · LCR 83.3 · MMMU-Pro 79.7
**Caveats**
@@ -323,7 +254,7 @@ MiMo Flash model for multimodal coding agents and long-context automation.
The cheapest multimodal agent in Go at 14¢/28¢ per 1M: bulk automation over long contexts, agents that must read images, video or audio, and the majority of mechanical coding-agent traffic.
*Evaluated:* Intelligence Index 37.9 · FrontierCode 42.1 · TB4.0 22.7 · AutomationBench 64.1 · HLE 35.1 · LCR 74.3 · MMMU-Pro 73.1
*Evaluated:* Intelligence Index 26.7 · FrontierCode 42.1 · TB4.0 22.7 · AutomationBench 64.1 · HLE 35.1 · LCR 74.3 · MMMU-Pro 73.1
**Caveats**
@@ -332,22 +263,6 @@ The cheapest multimodal agent in Go at 14¢/28¢ per 1M: bulk automation over lo
### undefined
**openai** · released 2026-07-09 · $0.2 in / $1.2 out per 1M tokens · 1.05M context · text, image, pdf in → text out
Targeted for: **Fast & cheap, High-volume coding, Agents**
Cost-efficient GPT-5.6 model for fast, high-volume workloads.
Throughput work: many small calls, classification, extraction, routing and repetitive edits where latency and cost per call decide the outcome. Keeps the GPT tool-calling behaviour at roughly twice the price of GPT-6 Luna.
*Evaluated:* Intelligence Index 37.3 · TB2.1 80.9 · AutomationBench 50.2 · GPQA 91.1 · HLE 39.5 · LCR 83.7 · MMMU-Pro 78.6
**Caveats**
- Price steps up above 272K tokens of context.
### undefined
**deepseek** · released 2026-04-24 · $0.66 in / $1.98 out per 1M tokens · 1M context · text in → text out
Targeted for: **Long agent runs, Coding, Math & science, Open weights**
@@ -356,7 +271,7 @@ Open MiE flagship with million-token context for coding and long agent runs.
The heavier DeepSeek: multi-hour agent sessions and research runs that need a large window, plus the maths and science scores (IMOAnswerBench, HMMT, LiveCodeBench) for technical work. Text-only, so it is a reasoning/code model rather than a perception model.
*Evaluated:* Intelligence Index 36 · FrontierCode 38.8 · TB2.1 78.7 · AutomationBench 56.7 · GPQA 92.8 · HLE 41 · LCR 80.3
*Evaluated:* Intelligence Index 23.1 · FrontierCode 38.8 · TB2.1 78.7 · AutomationBench 56.7 · GPQA 92.8 · HLE 41 · LCR 80.3
**Caveats**
@@ -365,40 +280,6 @@ The heavier DeepSeek: multi-hour agent sessions and research runs that need a la
### undefined
**zhipuai** · released 2026-06-13 · $1.4 in / $4.4 out per 1M tokens · 1M context · text in → text out
Targeted for: **Long-horizon coding, Agents, Math & science**
Open flagship GLM for long-horizon coding agents and million-token context work.
The GLM generation before 5.3: still strong on long-horizon coding agents with a million-token window, and its AIME/HMMT/IMO scores make it the maths-and-science option in the family. GLM-5.3 is the better default where budget allows.
*Evaluated:* Intelligence Index 33.7 · TB2.1 77.9 · AutomationBench 28.4 · GPQA 89.5 · HLE 41.1 · LCR 78.3
**Caveats**
- Text input only.
- Expensive per token ($1.40 / $4.40).
### undefined
**reflection** · released 2026-10-05 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1.05M context · text in → text out
Targeted for: **Coding, Reasoning, Long-horizon agents, Free tier**
Reflection's first open-weight model: a 501B-parameter sparse Mixture-of-Experts with 23B active per token, built for coding, reasoning and long-horizon agentic workloads.
Cost-efficient coding and agent loops. The MoE design activates only 23B parameters per token, so it is a cheap way to run a lot of terminal work, tool calls and multi-step reasoning. Text-only, so it is a pure code/reasoning worker rather than a perception model.
*Evaluated:* Intelligence Index 30.3 · FrontierCode 24.3 · TB4.0 3 · AutomationBench 48.5 · HLE 34.7 · LCR 78
**Caveats**
- Free for a limited time and unavailable in a long list of countries (see the OpenCode Go docs for the exclusion list).
- Weights are slated for an Apache 2.0 release, with the technical report and model card following.
### undefined
**minimax** · released 2026-06-01 · $0.3 in / $1.2 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Long-context coding, Perception, Agent planning, SWE tasks**
@@ -407,7 +288,7 @@ MiniMax multimodal model for long-context coding, perception, and agent planning
Repo-scale SWE work with perception in the loop: MiniMax reports one of the strongest SWE-Bench Verified results in Go, plus good browser and desktop-computer scores (BrowseComp, MCP Atlas, OSWorld). Use it when the agent has to look at a screen and then change code.
*Evaluated:* Intelligence Index 29.2 · FrontierCode 23 · TB2.1 65.2 · AutomationBench 21.3 · GPQA 92.9 · HLE 39 · LCR 83 · MMMU-Pro 78.6
*Evaluated:* Intelligence Index 18.7 · FrontierCode 23 · TB2.1 65.2 · AutomationBench 21.3 · GPQA 92.9 · HLE 39 · LCR 83 · MMMU-Pro 78.6
**Caveats**
@@ -416,54 +297,20 @@ Repo-scale SWE work with perception in the loop: MiniMax reports one of the stro
### undefined
**xiaomi** · released 2026-04-22 · $0.435 in / $0.87 out per 1M tokens · 1.05M context · text in → text out
**reflection** · released 2026-10-05 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1.05M context · text in → text out
Targeted for: **Multimodal coding agents, Long-context automation**
Targeted for: **Coding, Reasoning, Long-horizon agents, Free tier**
Previous-generation MiMo Pro tier for multimodal reasoning and coding-agent execution.
Reflection's first open-weight model: a 501B-parameter sparse Mixture-of-Experts with 23B active per token, built for coding, reasoning and long-horizon agentic workloads.
Now the value option in the MiMo line — same interface as 2.6 Pro with materially weaker agentic and cyber benchmarks. Reasonable when you need multimodal input cheaply and the task is not frontier-hard.
Cost-efficient coding and agent loops. The MoE design activates only 23B parameters per token, so it is a cheap way to run a lot of terminal work, tool calls and multi-step reasoning. Text-only, so it is a pure code/reasoning worker rather than a perception model.
*Evaluated:* Intelligence Index 26 · TB2.1 65.2 · AutomationBench 13.7 · GPQA 86.6 · HLE 35.7 · LCR 79.7
*Evaluated:* Intelligence Index 17.7 · FrontierCode 24.3 · TB4.0 3 · AutomationBench 48.5 · HLE 34.7 · LCR 78
**Caveats**
- Superseded by MiMo-V2.6-Pro; scores are materially lower on agentic and security tasks.
- Text output only.
### undefined
**moonshotai** · released 2026-06-12 · $0.95 in / $4 out per 1M tokens · 0.26M context · text, image, video in → text out
Targeted for: **Repo-scale coding, Long-horizon repo work, Tool use**
Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking.
Purpose-built for code: sustained repository work, refactors and multi-file changes where a general model wastes tokens deliberating. The tool-use (τ-bench) score is very high, so it is also good at driving MCP servers and CLI tools. Use it when the deliverable is a diff, not an essay.
*Evaluated:* Intelligence Index 25.8 · TB2.1 67.4 · AutomationBench 24.5 · GPQA 89.6 · HLE 35 · LCR 79.3
**Caveats**
- 256K context — roughly a quarter of the 1M models.
- Narrower general-purpose ability than the flagship Kimi.
### undefined
**tencent** · released 2026-07-06 · $0.14 in / $0.58 out per 1M tokens · 0.26M context · text in → text out
Targeted for: **Cheap, Reasoning, Instruction following, Tool use**
Tencent Hy reasoning model for coding, instruction following, and agent tasks.
A cheap reasoning model with unusually strong instruction following (IFBench) and tool use (τ-bench) for its price. Suited to spec-driven agents and short coding loops; the 256K window is the main limitation for repo-scale work.
*Evaluated:* Intelligence Index 25.3 · TB2.1 64.4 · AutomationBench 18.4 · GPQA 89.7 · HLE 33.5 · LCR 79
**Caveats**
- 256K context.
- Text input only.
- Free for a limited time and unavailable in a long list of countries (see the OpenCode Go docs for the exclusion list).
- Weights are slated for an Apache 2.0 release, with the technical report and model card following.
### undefined
@@ -475,7 +322,7 @@ Multimodal Qwen workhorse for long-context agents, visual inputs, and coding.
A steady middle tier: long-context agent sessions, image/video understanding and ordinary coding, without the flagship price. Instruction-following scores are its standout metric, which matters for agents driven by precise specs.
*Evaluated:* Intelligence Index 25.2 · TB2.1 61 · AutomationBench 17.4 · GPQA 90 · HLE 35.6 · LCR 73 · MMMU-Pro 80.5
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
@@ -484,74 +331,6 @@ A steady middle tier: long-context agent sessions, image/video understanding and
### undefined
**minimax** · released 2026-03-18 · $0.3 in / $1.2 out per 1M tokens · 0.20M context · text in → text out
Targeted for: **Coding, Office work, Agentic tasks**
MiniMax model for chat, coding, office work, and agentic tasks.
The generation before MiniMax M3: general chat plus coding and office deliverables. Cheap, but the 200K window and older agentic scores make it a fallback now that M3 and M3.1 Flash exist.
*Evaluated:* Intelligence Index 22.8 · TB2.1 55.4 · AutomationBench 4.4 · GPQA 87.4 · HLE 29.6 · LCR 78.3
**Caveats**
- 200K context — the smallest window in the Go catalogue.
- Superseded by MiniMax M3.
### undefined
**xiaomi** · released 2026-04-22 · $0.14 in / $0.28 out per 1M tokens · 1M context · text, image, audio, video in → text out
Targeted for: **Budget multimodal agents, Long context**
Open MiMo model for multimodal coding agents and long-context automation.
The cheapest way to get multimodal, long-context agent behaviour in Go: 14¢/28¢ per 1M. Best suited to bulk automation — many small steps over a large context — rather than to hard single-shot questions.
*Evaluated:* Intelligence Index 21.8 · TB2.1 63.7 · AutomationBench 18.4 · GPQA 84.9 · HLE 27.2 · LCR 73 · MMMU-Pro 75.4
**Caveats**
- Lowest tier of the MiMo line; weak on the hardest agentic benchmarks.
- Text output only.
### undefined
**meituan** · released 2026-06-30 · $0.3 in / $1.2 out per 1M tokens · 1M context · text in → text out
Targeted for: **Coding, Reasoning, Long context, Open weights**
Meituan LongCat-2.0, a reasoning model with tool calling and a 1M-token context window.
Long-context reasoning with tool calling at a modest price — useful for repo-scale analysis and document-heavy agent loops. Text-only, so perception work needs the 2.5 preview.
*Evaluated:* Intelligence Index 19.1 · TB2.1 50.2 · AutomationBench 4.5 · GPQA 78 · HLE 33.7 · LCR 65
**Caveats**
- Text input only.
- Superseded by LongCat 2.5 Preview, which adds image input.
### undefined
**undisclosed** · released 2026-10-12 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1.05M context · text in → text out
Targeted for: **Free tier, Coding, Agentic workflows, Structured output**
A text-only reasoning model for coding and agentic workflows, with tool calling and structured JSON outputs.
An anonymous free tier aimed at agents: tool calling and structured JSON output make it useful as a sub-agent or a formatter that has to hand machine-readable results to the next step.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- No vendor, no model card, no public evaluation — verify behaviour before depending on it.
- Not listed in the live Go catalogue endpoint.
### undefined
**stepfun** · released 2026-09-16 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1M context · text, image, video in → text out
Targeted for: **Free tier, Coding, Professional knowledge work, Multimodal**
@@ -585,21 +364,149 @@ Productivity and workflow automation: multi-step office/enterprise tasks, tool o
### undefined
**meituan** · released 2026-09-25 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1M context · text, image in → text out
**undisclosed** · released 2026-10-12 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1.05M context · text in → text out
Targeted for: **Coding, Agentic, Image understanding, Free tier**
Targeted for: **Free tier, Coding, Agentic workflows, Structured output**
Meituan's multimodal reasoning model for coding and agent tasks, with image understanding and a 1M-token context window.
A text-only reasoning model for coding and agentic workflows, with tool calling and structured JSON outputs.
Agents that have to read a picture to act: UI automation, screenshot-to-code, chart and document understanding, then long runs of tool use against a 1M-token window. Pairs well with a free-tier workflow — long exploratory sessions without burning a paid model's allowance.
An anonymous free tier aimed at agents: tool calling and structured JSON output make it useful as a sub-agent or a formatter that has to hand machine-readable results to the next step.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- No vendor, no model card, no public evaluation — verify behaviour before depending on it.
- Not listed in the live Go catalogue endpoint.
### undefined
**alibaba** · released 2026-08-26 · $0.15 in / $0.47 out per 1M tokens · 1M context · text, image, video in → text out
Targeted for: **Visual reasoning, Documents, Agent tasks, Cheap**
Qwen vision-language model for visual reasoning, documents, and agent tasks.
Perception-heavy work at a low price: reading screenshots and diagrams, extracting from documents, and driving agents that need to see what is on screen. It pairs a 1M window with sub-20¢ input pricing, which makes it a good bulk worker for visual pipelines.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Free for a limited time.
- Preview release: scores and behaviour may shift.
- Text output only — no image generation.
- Evaluated publicly as 'Qwen3.8-Flash-Next' in third-party benchmark tables.
### undefined
**moonshotai** · released 2026-07-16 · $3 in / $15 out per 1M tokens · 1.05M context · text, image, video in → text out
Targeted for: **Long-horizon agents, Coding, Web research, Spreadsheets, Open weights**
Multimodal Kimi model with a 1M context and toggleable max-effort thinking for long-horizon agent work.
The Kimi tier built for endurance: long-horizon agent runs, web research (BrowseComp), coding (FrontierSWE, DeepSWE, Terminal-Bench), and office deliverables — it scores well on SpreadsheetBench, which makes it the pick for data-analysis and spreadsheet work in Go. Thinking effort can be turned up for hard problems or down for speed.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Most expensive model in Go at $3 / $15 per 1M tokens.
- Terminal-Bench 4.0 is much weaker than its Terminal-Bench 2.1 result — hard-shell tasks still cost it a lot.
- Text output only.
### undefined
**tencent** · released 2026-07-06 · $0.14 in / $0.58 out per 1M tokens · 0.26M context · text in → text out
Targeted for: **Cheap, Reasoning, Instruction following, Tool use**
Tencent Hy reasoning model for coding, instruction following, and agent tasks.
A cheap reasoning model with unusually strong instruction following (IFBench) and tool use (τ-bench) for its price. Suited to spec-driven agents and short coding loops; the 256K window is the main limitation for repo-scale work.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- 256K context.
- Text input only.
### undefined
**deepseek** · released 2026-09-10 · $0.15 in / $0.6 out per 1M tokens · 1M context · text, image in → text out
Targeted for: **Agentic coding, Reasoning, Cheap, High volume**
DeepSeek's V4.1 Flash: built for reasoning and agentic coding, and the single most-used model in OpenCode by token volume.
The default engine for long coding-agent runs. DeepSeek's own card lists Terminal-Bench, DeepSWE, CyberGym and AutomationBench scores, i.e. it is tuned for terminal work, repo-level fixes and security tasks rather than chat. With cache reads at ~0.3¢/1M it is also the cheapest way to move large volumes of agent traffic.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Text output only.
- Price doubles during DeepSeek's peak hours (01:00–04:00 and 06:00–10:00 UTC, Mon–Fri).
### undefined
**undisclosed** · released 2026-10-24 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 0.26M context · text in → text out
Stealth model, free for a limited time
_No hand-written write-up yet._
*Evaluated:* no public third-party evaluation published yet.
### undefined
**openai** · released 2026-07-09 · $0.2 in / $1.2 out per 1M tokens · 1.05M context · text, image, pdf in → text out
Targeted for: **Fast & cheap, High-volume coding, Agents**
Cost-efficient GPT-5.6 model for fast, high-volume workloads.
Throughput work: many small calls, classification, extraction, routing and repetitive edits where latency and cost per call decide the outcome. Keeps the GPT tool-calling behaviour at roughly twice the price of GPT-6 Luna.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Price steps up above 272K tokens of context.
### undefined
**moonshotai** · released 2026-06-12 · $0.95 in / $4 out per 1M tokens · 0.26M context · text, image, video in → text out
Targeted for: **Repo-scale coding, Long-horizon repo work, Tool use**
Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking.
Purpose-built for code: sustained repository work, refactors and multi-file changes where a general model wastes tokens deliberating. The tool-use (τ-bench) score is very high, so it is also good at driving MCP servers and CLI tools. Use it when the deliverable is a diff, not an essay.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- 256K context — roughly a quarter of the 1M models.
- Narrower general-purpose ability than the flagship Kimi.
### undefined
**xiaomi** · released 2026-04-22 · $0.14 in / $0.28 out per 1M tokens · 1M context · text, image, audio, video in → text out
Targeted for: **Budget multimodal agents, Long context**
Open MiMo model for multimodal coding agents and long-context automation.
The cheapest way to get multimodal, long-context agent behaviour in Go: 14¢/28¢ per 1M. Best suited to bulk automation — many small steps over a large context — rather than to hard single-shot questions.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Lowest tier of the MiMo line; weak on the hardest agentic benchmarks.
- Text output only.
### undefined
@@ -620,14 +527,107 @@ An unreleased preview model with full multimodal input and tool use, priced like
### undefined
**undisclosed** · released 2026-10-24 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 0.26M context · text in → text out
**meituan** · released 2026-09-25 · $0 in / $0 out per 1M tokens · **free tier (limited time)** · 1M context · text, image in → text out
Stealth model, free for a limited time
Targeted for: **Coding, Agentic, Image understanding, Free tier**
_No hand-written write-up yet._
Meituan's multimodal reasoning model for coding and agent tasks, with image understanding and a 1M-token context window.
Agents that have to read a picture to act: UI automation, screenshot-to-code, chart and document understanding, then long runs of tool use against a 1M-token window. Pairs well with a free-tier workflow — long exploratory sessions without burning a paid model's allowance.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Free for a limited time.
- Preview release: scores and behaviour may shift.
- Text output only — no image generation.
### undefined
**zhipuai** · released 2026-06-13 · $1.4 in / $4.4 out per 1M tokens · 1M context · text in → text out
Targeted for: **Long-horizon coding, Agents, Math & science**
Open flagship GLM for long-horizon coding agents and million-token context work.
The GLM generation before 5.3: still strong on long-horizon coding agents with a million-token window, and its AIME/HMMT/IMO scores make it the maths-and-science option in the family. GLM-5.3 is the better default where budget allows.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Text input only.
- Expensive per token ($1.40 / $4.40).
### undefined
**xiaomi** · released 2026-04-22 · $0.435 in / $0.87 out per 1M tokens · 1.05M context · text in → text out
Targeted for: **Multimodal coding agents, Long-context automation**
Previous-generation MiMo Pro tier for multimodal reasoning and coding-agent execution.
Now the value option in the MiMo line — same interface as 2.6 Pro with materially weaker agentic and cyber benchmarks. Reasonable when you need multimodal input cheaply and the task is not frontier-hard.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Superseded by MiMo-V2.6-Pro; scores are materially lower on agentic and security tasks.
- Text output only.
### undefined
**minimax** · released 2026-03-18 · $0.3 in / $1.2 out per 1M tokens · 0.20M context · text in → text out
Targeted for: **Coding, Office work, Agentic tasks**
MiniMax model for chat, coding, office work, and agentic tasks.
The generation before MiniMax M3: general chat plus coding and office deliverables. Cheap, but the 200K window and older agentic scores make it a fallback now that M3 and M3.1 Flash exist.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- 200K context — the smallest window in the Go catalogue.
- Superseded by MiniMax M3.
### undefined
**meituan** · released 2026-06-30 · $0.3 in / $1.2 out per 1M tokens · 1M context · text in → text out
Targeted for: **Coding, Reasoning, Long context, Open weights**
Meituan LongCat-2.0, a reasoning model with tool calling and a 1M-token context window.
Long-context reasoning with tool calling at a modest price — useful for repo-scale analysis and document-heavy agent loops. Text-only, so perception work needs the 2.5 preview.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Text input only.
- Superseded by LongCat 2.5 Preview, which adds image input.
### undefined
**meta** · released 2026-08-05 · $0.1 in / $0.2 out per 1M tokens · 1.05M context · text, image, video, pdf, audio in → text out
Targeted for: **Coding, Multimodal, Cheap**
Coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.
The previous Muse generation — kept in Go because it is cheap and coding-focused. Pick it when you want Meta's multimodal input surface at $0.10/$0.20 and 1.3 is not required; otherwise use Muse Spark 1.3.
*Evaluated:* no public third-party evaluation published yet.
**Caveats**
- Contributor tier — prompts and completions may be used to train future Meta models.
- Superseded by Muse Spark 1.3 on long-horizon agent tasks.
## Free models outside the Go plan
### MiMo-V2.6-Flash Free (Xiaomi)