OpenCode Models Comparison
1 July 2026 | Eric
The goal of this article is to answer the question: “Which model should I use for best cost-efficiency?”. The reality of today’s AI landscape which we already explored is that some models are only good because they are highly subsidized behind opaque subscriptions. But for you and me using token-based harnesses, like OpenCode, there are real decisions to be made. Below is a simple heuristic to make these decisions.
Choosing a Model for a Task
Calculating model efficiency is difficult. There are many parameters, like difficulty of the task, token-efficiency, token cost, caching rate, providers, etc. In the model below, I avoided collapsing cost and capability into a single lossy index, instead the practical approach is a filter-then-rank decision procedure:
- Threshold: Set minimum capability needed for the task.
- Filter: Filter-out under-qualified models.
- Rank: Sort by Effective Cost ascending; ties broken by Intelligence Index, then Coding Index, then Agentic Index (higher is better).
- Pick: Cheapest model is our best value.
What is Effective Cost?
Effective Cost is a blended per-million-token rate assuming a default 80% cache hit rate and a 3:1 input:output token ratio (75% input, 25% output). Formula: 0.75 × (0.20 × Input + 0.80 × Cached Read) + 0.25 × Output. Cache write pricing is excluded as it’s a one-time cost per unique prompt prefix, not a per-request cost.
Provider: OpenCode Zen pricing as of July 2026. Index scores sourced from Artificial Analysis Intelligence Index v4.1, Coding Index, and Agentic Index.
AA Cost Ratio is the model’s total cost to evaluate on the AA Intelligence Index ÷ Claude Opus 4.8’s total cost ($3,752.55). Lower is better. Baseline (Claude Opus 4.8) set to 1.000.
Free models are excluded. Some models like Gemini 3.1 and GPT 5.x have similar scores disregarding the context window tier.
Use the sliders below to set minimum Intelligence, Coding, and Agentic scores. The table updates instantly and ranks surviving models by Effective Cost ascending, with ties broken by Intelligence, then Coding, then Agentic index (higher is better).
Loading models…
How to calibrate your own threshold
But before you do that, read below about Cost Staircase & Capability Tiers.
| Task type | Primary index | Typical threshold |
|---|---|---|
| Simple classification, keyword extraction | Intelligence | 30–35 |
| Structured data extraction, summarization | Intelligence | 35–40 |
| RAG response generation, Q&A | Intelligence | 40–50 |
| Complex reasoning, deal analysis, proposals | Intelligence | 50–55 |
| Top-tier research, ambiguous problem-solving | Intelligence | 55+ |
| Simple codegen (boilerplate, scripts, auto-completion, inline assist) | Coding | 45–55 |
| Feature-level codegen, bug fixing | Coding | 55–65 |
| Complex codegen (multi-file, architecture) | Coding | 65+ |
| Tool-calling, single-step agents | Agentic | 25–35 |
| Multi-step agents, error recovery | Agentic | 35–45 |
| Autonomous long-running agents | Agentic | 45+ |
Some observations
GPT 5.6 Luna
On July 30, 2026, OpenAI cut GPT 5.6 Luna pricing by about 80% and GPT 5.6 Terra by about 20%. That single price move rewrote the middle of the table.
Luna (≤ 272K) now sits at an Effective Cost of $0.34 per million blended tokens. Against models with near-parity Coding and Agentic scores:
- Claude Sonnet 5 ($2.92) is about 8.5× more expensive for +2 Intelligence, +0.1 Coding, +1.1 Agentic
- Grok 4.5 (≤ 200K) ($2.10) is about 6× more expensive for +3 Intelligence, +1.0 Coding, +0.1 Agentic
MiniMax M3, Kimi K2.7 Code, and GLM 5.2 were the natural mid-tier picks a week ago. Luna undercuts all three on Effective Cost while clearing their Coding and Agentic bars in one step. Grok 4.5 and Sonnet 5 still edge Luna on Intelligence, but the premium is hard to justify when the capability delta is a few index points.
Claude Sonnet 5
Claude Sonnet 5 remains expensive on another axis: it consumed more tokens than any other model on the AA benchmark, which pushed its AA Cost Ratio to 1.6. It is even more expensive than Claude Fable 5 on that measure while being less efficient. I don’t really know yet how to justify this. Maybe it is optimized for the Anthropic toolchain only. After the Luna cut, the gap is even harder to ignore.
Cost Staircase & Capability Tiers
What we’re seeing is a cost staircase. The table looks like two dozen individual models, but it is really a handful of capability tiers with large price steps between them. The jumps happen at specific coding thresholds where the cheap model falls off. Those thresholds are where model selection matters.
After the Luna cut, the middle rungs disappeared. MiniMax M3, Kimi K2.7, and GLM 5.2 no longer define steps on the coding path. Luna spans what used to be three tiers.
| Coding threshold | Cheapest survivor | Cost | Jump from previous |
|---|---|---|---|
| < 57 | DeepSeek V4 Flash | $0.11 | — (baseline) |
| 57–71 | GPT 5.6 Luna (≤ 272K) | $0.34 | 3.2× |
| 72 | Grok 4.5 (≤ 200K) | $2.10 | 6.1× |
| 73–76 | GPT 5.6 Terra (≤ 272K) | $3.42 | 1.6× |
| ≥ 77 | Claude Opus 5 | $7.30 | 2.1× |
The intelligence path shows the same pattern, with smaller score gaps and large cost jumps:
- 40 → 41 (+1 pt): 3.2×: DeepSeek V4 Flash drops off, Luna takes over
- 51 → 52 (+1 pt): 6.1×: Luna drops off, Grok 4.5 takes over
- 54 → 55 (+1 pt): 1.6×: Grok drops off, Terra takes over
- 57 → 58 (+1 pt): 1.7×: Kimi K3 drops off, Claude Opus 5 takes over
A single intelligence point past Luna still costs you about 6× more per token. That is the cliff that matters now.
Changelog
- 2026-07-09 — Added Claude Fable 5 to the model comparison table.
- 2026-07-09 — Added Grok 4.5 (≤ 200K and > 200K tiers) to the model comparison table.
- 2026-07-13 — Added GPT 5.6 Sol, Terra, and Luna (≤ 272K and > 272K tiers) to the model comparison table.
- 2026-07-25 — Added Claude Opus 5 to the model comparison table.
- 2026-07-28 — Added Kimi K3 to the model comparison table.
- 2026-07-28 — Added Gemini 3.6 Flash to the model comparison table.
- 2026-07-31 — Updated GPT 5.6 Terra and Luna pricing (Sol unchanged).
- 2026-07-31 — Refreshed observations and cost staircase after the Luna/Terra price cut.
- 2026-08-12 — Removed the Artificial Analysis Coding Index from the comparison table.
- 2026-08-15 — Added Gemini 3.7 Flash, Grok 4.6 (≤ 200K and > 200K tiers), and Muse Spark 1.2 to the model comparison table.